webui: fix code copy stripping XML/HTML tags (#18518 )

* webui: fix code copy stripping XML/HTML tags * webui: update static build
ggml-cuda: remove unneccesary prints on ggml_cuda_init (#18502 )
2026-01-01 13:44:11 +01:00 · 2026-01-01 19:18:43 +08:00 · 2026-01-01 08:58:27 +01:00 · 2026-01-01 10:24:43 +08:00 · 2025-12-31 17:21:37 -06:00 · 2025-12-31 18:54:43 +02:00
31 changed files with 917 additions and 426 deletions
--- a/common/chat.cpp
+++ b/common/chat.cpp
@ -380,8 +380,8 @@ std::vector<common_chat_tool> common_chat_tools_parse_oaicompat(const json & too
                const auto & function = tool.at("function");
                result.push_back({
                    /* .name = */ function.at("name"),
-                    /* .description = */ function.at("description"),
-                    /* .parameters = */ function.at("parameters").dump(),
+                    /* .description = */ function.value("description", ""),
+                    /* .parameters = */ function.value("parameters", json::object()).dump(),
                });
            }
        }
--- a/convert_hf_to_gguf.py
+++ b/convert_hf_to_gguf.py
@ -3503,7 +3503,7 @@ class QwenModel(TextModel):
        self._set_vocab_qwen()


-@ModelBase.register("Qwen2Model", "Qwen2ForCausalLM", "Qwen2AudioForConditionalGeneration", "KORMoForCausalLM")
+@ModelBase.register("Qwen2Model", "Qwen2ForCausalLM", "Qwen2AudioForConditionalGeneration", "KORMoForCausalLM", "AudioFlamingo3ForConditionalGeneration")
 class Qwen2Model(TextModel):
    model_arch = gguf.MODEL_ARCH.QWEN2

@ -9292,6 +9292,19 @@ class VoxtralWhisperEncoderModel(WhisperEncoderModel):
        self.gguf_writer.add_audio_stack_factor(4) # == intermediate_size // hidden_size


+@ModelBase.register("AudioFlamingo3ForConditionalGeneration")
+class AudioFlamingo3WhisperEncoderModel(WhisperEncoderModel):
+    def set_gguf_parameters(self):
+        super().set_gguf_parameters()
+        self.gguf_writer.add_clip_projector_type(gguf.VisionProjectorType.MUSIC_FLAMINGO)
+
+    def tensor_force_quant(self, name, new_name, bid, n_dims):
+        if ".conv" in name and ".weight" in name:
+            # Was trained in BF16, being safe, avoiding quantizing to FP16
+            return gguf.GGMLQuantizationType.F32
+        return super().tensor_force_quant(name, new_name, bid, n_dims)
+
+
@ModelBase.register("FalconH1ForCausalLM")
 class FalconH1Model(Mamba2Model):
    model_arch = gguf.MODEL_ARCH.FALCON_H1
--- a/docs/ops.md
+++ b/docs/ops.md
@ -32,7 +32,7 @@ Legend:
 |                CONV_TRANSPOSE_1D | ❌ | ✅ | ✅ | ✅ | ✅ | ❌ | ✅ | ✅ | ❌ | ❌ | ❌ |
 |                CONV_TRANSPOSE_2D | ❌ | ❌ | ✅ | ✅ | ✅ | ❌ | ❌ | ✅ | ❌ | ❌ | ❌ |
 |                              COS | ❌ | ✅ | ✅ | ✅ | 🟡 | ❌ | ✅ | 🟡 | ❌ | ❌ | ❌ |
-|                      COUNT_EQUAL | ❌ | ✅ | ✅ | ✅ | ❌ | ❌ | ✅ | ✅ | ❌ | ❌ | ❌ |
+|                      COUNT_EQUAL | ❌ | ✅ | ✅ | ✅ | ✅ | ❌ | ✅ | ✅ | ❌ | ❌ | ❌ |
 |                              CPY | ❌ | 🟡 | 🟡 | 🟡 | 🟡 | 🟡 | 🟡 | 🟡 | 🟡 | ❌ | ❌ |
 |               CROSS_ENTROPY_LOSS | ❌ | ❌ | ✅ | ✅ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ |
 |          CROSS_ENTROPY_LOSS_BACK | ❌ | ❌ | ✅ | ✅ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ |
--- a/docs/ops/Metal.csv
+++ b/docs/ops/Metal.csv
@ -965,6 +965,7 @@
 "Metal","IM2COL","type_input=f32,type_kernel=f16,dst_type=f16,ne_input=[12,12,1,2560],ne_kernel=[3,3,1,2560],s0=1,s1=1,p0=1,p1=1,d0=1,d1=1,is_2D=1","support","1","yes","Metal"
 "Metal","IM2COL","type_input=f32,type_kernel=f16,dst_type=f16,ne_input=[12,12,2,2560],ne_kernel=[3,3,2,2560],s0=1,s1=1,p0=1,p1=1,d0=1,d1=1,is_2D=1","support","1","yes","Metal"
 "Metal","IM2COL","type_input=f32,type_kernel=f16,dst_type=f16,ne_input=[5,5,1,32],ne_kernel=[3,4,1,32],s0=1,s1=1,p0=0,p1=0,d0=1,d1=1,is_2D=1","support","1","yes","Metal"
+"Metal","IM2COL","type_input=f32,type_kernel=f32,dst_type=f32,ne_input=[2,2,1536,729],ne_kernel=[2,2,1536,4096],s0=1,s1=1,p0=0,p1=0,d0=1,d1=1,is_2D=1","support","1","yes","Metal"
 "Metal","IM2COL_3D","type_input=f32,type_kernel=f32,dst_type=f32,ne_input=[10,10,10,9],ne_kernel=[3,3,3,1],IC=3,s0=1,s1=1,s2=1,p0=1,p1=1,p2=1,d0=1,d1=1,d2=1,v=0","support","0","no","Metal"
 "Metal","IM2COL_3D","type_input=f32,type_kernel=f16,dst_type=f32,ne_input=[10,10,10,9],ne_kernel=[3,3,3,1],IC=3,s0=1,s1=1,s2=1,p0=1,p1=1,p2=1,d0=1,d1=1,d2=1,v=0","support","0","no","Metal"
 "Metal","IM2COL_3D","type_input=f32,type_kernel=f16,dst_type=f16,ne_input=[10,10,10,9],ne_kernel=[3,3,3,1],IC=3,s0=1,s1=1,s2=1,p0=1,p1=1,p2=1,d0=1,d1=1,d2=1,v=0","support","0","no","Metal"
@ -4964,8 +4965,9 @@
 "Metal","CONV_TRANSPOSE_1D","ne_input=[2,1,1,1],ne_kernel=[3,1,1,1],s0=1,p0=0,d0=1","support","1","yes","Metal"
 "Metal","CONV_TRANSPOSE_2D","ne_input=[3,2,3,1],ne_kernel=[2,2,1,3],stride=1","support","1","yes","Metal"
 "Metal","CONV_TRANSPOSE_2D","ne_input=[10,10,9,1],ne_kernel=[3,3,1,9],stride=2","support","1","yes","Metal"
-"Metal","COUNT_EQUAL","type=f32,ne=[4,500,1,1]","support","0","no","Metal"
-"Metal","COUNT_EQUAL","type=f32,ne=[4,5000,1,1]","support","0","no","Metal"
+"Metal","CONV_TRANSPOSE_2D","ne_input=[129,63,35,1],ne_kernel=[3,3,48,35],stride=1","support","1","yes","Metal"
+"Metal","COUNT_EQUAL","type=f32,ne=[4,500,1,1]","support","1","yes","Metal"
+"Metal","COUNT_EQUAL","type=f32,ne=[4,5000,1,1]","support","1","yes","Metal"
 "Metal","ARGMAX","type=f32,ne=[32,1,1,1]","support","1","yes","Metal"
 "Metal","ARGMAX","type=f32,ne=[32,513,1,1]","support","1","yes","Metal"
 "Metal","ARGMAX","type=f32,ne=[100,10,1,1]","support","1","yes","Metal"
@ -5715,15 +5717,15 @@
 "Metal","L2_NORM","type=f32,ne=[64,5,4,3]","support","1","yes","Metal"
 "Metal","RMS_NORM","type=f32,ne=[64,5,4,3],v=0,eps=0.000001,inplace=1","support","1","yes","Metal"
 "Metal","L2_NORM","type=f32,ne=[64,5,4,3]","support","1","yes","Metal"
-"Metal","SSM_CONV","type=f32,ne_a=[4,1024,1,1],ne_b=[3,1024,1,1]","support","1","yes","Metal"
-"Metal","SSM_CONV","type=f32,ne_a=[8,1024,1,1],ne_b=[3,1024,1,1]","support","1","yes","Metal"
-"Metal","SSM_CONV","type=f32,ne_a=[4,1024,4,1],ne_b=[3,1024,1,1]","support","1","yes","Metal"
-"Metal","SSM_CONV","type=f32,ne_a=[4,1536,1,1],ne_b=[3,1536,1,1]","support","1","yes","Metal"
-"Metal","SSM_CONV","type=f32,ne_a=[8,1536,1,1],ne_b=[3,1536,1,1]","support","1","yes","Metal"
-"Metal","SSM_CONV","type=f32,ne_a=[4,1536,4,1],ne_b=[3,1536,1,1]","support","1","yes","Metal"
-"Metal","SSM_CONV","type=f32,ne_a=[4,2048,1,1],ne_b=[3,2048,1,1]","support","1","yes","Metal"
-"Metal","SSM_CONV","type=f32,ne_a=[8,2048,1,1],ne_b=[3,2048,1,1]","support","1","yes","Metal"
-"Metal","SSM_CONV","type=f32,ne_a=[4,2048,4,1],ne_b=[3,2048,1,1]","support","1","yes","Metal"
+"Metal","SSM_CONV","type=f32,ne_a=[3,1024,1,1],ne_b=[3,1024,1,1]","support","1","yes","Metal"
+"Metal","SSM_CONV","type=f32,ne_a=[6,1024,1,1],ne_b=[3,1024,1,1]","support","1","yes","Metal"
+"Metal","SSM_CONV","type=f32,ne_a=[3,1024,4,1],ne_b=[3,1024,1,1]","support","1","yes","Metal"
+"Metal","SSM_CONV","type=f32,ne_a=[3,1536,1,1],ne_b=[3,1536,1,1]","support","1","yes","Metal"
+"Metal","SSM_CONV","type=f32,ne_a=[6,1536,1,1],ne_b=[3,1536,1,1]","support","1","yes","Metal"
+"Metal","SSM_CONV","type=f32,ne_a=[3,1536,4,1],ne_b=[3,1536,1,1]","support","1","yes","Metal"
+"Metal","SSM_CONV","type=f32,ne_a=[3,2048,1,1],ne_b=[3,2048,1,1]","support","1","yes","Metal"
+"Metal","SSM_CONV","type=f32,ne_a=[6,2048,1,1],ne_b=[3,2048,1,1]","support","1","yes","Metal"
+"Metal","SSM_CONV","type=f32,ne_a=[3,2048,4,1],ne_b=[3,2048,1,1]","support","1","yes","Metal"
 "Metal","SSM_CONV","type=f32,ne_a=[4,1024,1,1],ne_b=[4,1024,1,1]","support","1","yes","Metal"
 "Metal","SSM_CONV","type=f32,ne_a=[8,1024,1,1],ne_b=[4,1024,1,1]","support","1","yes","Metal"
 "Metal","SSM_CONV","type=f32,ne_a=[4,1024,4,1],ne_b=[4,1024,1,1]","support","1","yes","Metal"
@ -5733,6 +5735,15 @@
 "Metal","SSM_CONV","type=f32,ne_a=[4,2048,1,1],ne_b=[4,2048,1,1]","support","1","yes","Metal"
 "Metal","SSM_CONV","type=f32,ne_a=[8,2048,1,1],ne_b=[4,2048,1,1]","support","1","yes","Metal"
 "Metal","SSM_CONV","type=f32,ne_a=[4,2048,4,1],ne_b=[4,2048,1,1]","support","1","yes","Metal"
+"Metal","SSM_CONV","type=f32,ne_a=[9,1024,1,1],ne_b=[9,1024,1,1]","support","1","yes","Metal"
+"Metal","SSM_CONV","type=f32,ne_a=[18,1024,1,1],ne_b=[9,1024,1,1]","support","1","yes","Metal"
+"Metal","SSM_CONV","type=f32,ne_a=[9,1024,4,1],ne_b=[9,1024,1,1]","support","1","yes","Metal"
+"Metal","SSM_CONV","type=f32,ne_a=[9,1536,1,1],ne_b=[9,1536,1,1]","support","1","yes","Metal"
+"Metal","SSM_CONV","type=f32,ne_a=[18,1536,1,1],ne_b=[9,1536,1,1]","support","1","yes","Metal"
+"Metal","SSM_CONV","type=f32,ne_a=[9,1536,4,1],ne_b=[9,1536,1,1]","support","1","yes","Metal"
+"Metal","SSM_CONV","type=f32,ne_a=[9,2048,1,1],ne_b=[9,2048,1,1]","support","1","yes","Metal"
+"Metal","SSM_CONV","type=f32,ne_a=[18,2048,1,1],ne_b=[9,2048,1,1]","support","1","yes","Metal"
+"Metal","SSM_CONV","type=f32,ne_a=[9,2048,4,1],ne_b=[9,2048,1,1]","support","1","yes","Metal"
 "Metal","SSM_SCAN","type=f32,d_state=16,head_dim=1,n_head=1024,n_group=1,n_seq_tokens=32,n_seqs=4","support","1","yes","Metal"
 "Metal","SSM_SCAN","type=f32,d_state=128,head_dim=64,n_head=16,n_group=2,n_seq_tokens=32,n_seqs=4","support","1","yes","Metal"
 "Metal","SSM_SCAN","type=f32,d_state=256,head_dim=64,n_head=8,n_group=2,n_seq_tokens=32,n_seqs=4","support","1","yes","Metal"
@ -8916,6 +8927,8 @@
 "Metal","SOFT_MAX","type=f32,ne=[32,2,32,1],mask=1,sinks=0,m_prec=f16,nr23=[1,1],scale=0.100000,max_bias=0.000000,inplace=0","support","1","yes","Metal"
 "Metal","SOFT_MAX","type=f32,ne=[32,2,32,1],mask=1,sinks=1,m_prec=f32,nr23=[1,1],scale=0.100000,max_bias=8.000000,inplace=0","support","1","yes","Metal"
 "Metal","SOFT_MAX","type=f32,ne=[32,2,32,1],mask=1,sinks=1,m_prec=f16,nr23=[1,1],scale=0.100000,max_bias=8.000000,inplace=0","support","1","yes","Metal"
+"Metal","SOFT_MAX","type=f32,ne=[200001,2,3,1],mask=1,sinks=1,m_prec=f32,nr23=[1,1],scale=0.100000,max_bias=8.000000,inplace=0","support","1","yes","Metal"
+"Metal","SOFT_MAX","type=f32,ne=[200001,2,3,1],mask=1,sinks=1,m_prec=f16,nr23=[1,1],scale=0.100000,max_bias=8.000000,inplace=0","support","1","yes","Metal"
 "Metal","SOFT_MAX_BACK","type=f32,ne=[16,16,1,1],scale=1.000000,max_bias=0.000000","support","0","no","Metal"
 "Metal","SOFT_MAX_BACK","type=f32,ne=[15,15,1,1],scale=1.000000,max_bias=0.000000","support","0","no","Metal"
 "Metal","SOFT_MAX_BACK","type=f32,ne=[16,16,2,3],scale=1.000000,max_bias=0.000000","support","0","no","Metal"
@ -9542,311 +9555,311 @@
 "Metal","ARGSORT","type=f32,ne=[2048,2,1,3],order=1","support","1","yes","Metal"
 "Metal","ARGSORT","type=f32,ne=[2049,2,1,3],order=1","support","1","yes","Metal"
 "Metal","ARGSORT","type=f32,ne=[2,8,8192,1],order=1","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[1,1,1,1],k=1","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[12,1,2,1],k=1","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[2,1,1,1],k=1","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[13,1,2,1],k=1","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[2,1,1,1],k=2","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[13,1,2,1],k=2","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[4,1,1,1],k=1","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[15,1,2,1],k=1","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[4,1,1,1],k=2","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[15,1,2,1],k=2","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[4,1,1,1],k=3","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[15,1,2,1],k=3","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[8,1,1,1],k=1","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[19,1,2,1],k=1","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[8,1,1,1],k=2","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[19,1,2,1],k=2","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[8,1,1,1],k=3","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[19,1,2,1],k=3","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[8,1,1,1],k=7","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[19,1,2,1],k=7","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[16,1,1,1],k=1","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[27,1,2,1],k=1","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[16,1,1,1],k=2","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[27,1,2,1],k=2","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[16,1,1,1],k=3","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[27,1,2,1],k=3","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[16,1,1,1],k=7","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[27,1,2,1],k=7","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[16,1,1,1],k=15","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[27,1,2,1],k=15","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[32,1,1,1],k=1","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[43,1,2,1],k=1","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[32,1,1,1],k=2","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[43,1,2,1],k=2","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[32,1,1,1],k=3","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[43,1,2,1],k=3","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[32,1,1,1],k=7","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[43,1,2,1],k=7","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[32,1,1,1],k=15","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[43,1,2,1],k=15","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[64,1,1,1],k=1","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[75,1,2,1],k=1","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[64,1,1,1],k=2","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[75,1,2,1],k=2","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[64,1,1,1],k=3","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[75,1,2,1],k=3","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[64,1,1,1],k=7","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[75,1,2,1],k=7","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[64,1,1,1],k=15","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[75,1,2,1],k=15","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[128,1,1,1],k=1","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[139,1,2,1],k=1","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[128,1,1,1],k=2","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[139,1,2,1],k=2","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[128,1,1,1],k=3","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[139,1,2,1],k=3","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[128,1,1,1],k=7","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[139,1,2,1],k=7","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[128,1,1,1],k=15","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[139,1,2,1],k=15","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[128,1,1,1],k=100","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[139,1,2,1],k=100","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[256,1,1,1],k=1","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[267,1,2,1],k=1","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[256,1,1,1],k=2","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[267,1,2,1],k=2","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[256,1,1,1],k=3","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[267,1,2,1],k=3","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[256,1,1,1],k=7","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[267,1,2,1],k=7","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[256,1,1,1],k=15","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[267,1,2,1],k=15","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[256,1,1,1],k=100","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[267,1,2,1],k=100","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[512,1,1,1],k=1","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[523,1,2,1],k=1","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[512,1,1,1],k=2","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[523,1,2,1],k=2","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[512,1,1,1],k=3","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[523,1,2,1],k=3","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[512,1,1,1],k=7","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[523,1,2,1],k=7","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[512,1,1,1],k=15","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[523,1,2,1],k=15","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[512,1,1,1],k=100","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[523,1,2,1],k=100","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[512,1,1,1],k=500","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[523,1,2,1],k=500","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[1024,1,1,1],k=1","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[1035,1,2,1],k=1","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[1024,1,1,1],k=2","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[1035,1,2,1],k=2","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[1024,1,1,1],k=3","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[1035,1,2,1],k=3","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[1024,1,1,1],k=7","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[1035,1,2,1],k=7","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[1024,1,1,1],k=15","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[1035,1,2,1],k=15","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[1024,1,1,1],k=100","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[1035,1,2,1],k=100","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[1024,1,1,1],k=500","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[1035,1,2,1],k=500","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[1024,1,1,1],k=1023","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[1035,1,2,1],k=1023","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[2048,1,1,1],k=1","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[2059,1,2,1],k=1","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[2048,1,1,1],k=2","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[2059,1,2,1],k=2","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[2048,1,1,1],k=3","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[2059,1,2,1],k=3","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[2048,1,1,1],k=7","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[2059,1,2,1],k=7","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[2048,1,1,1],k=15","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[2059,1,2,1],k=15","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[2048,1,1,1],k=100","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[2059,1,2,1],k=100","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[2048,1,1,1],k=500","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[2059,1,2,1],k=500","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[2048,1,1,1],k=1023","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[2059,1,2,1],k=1023","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[4096,1,1,1],k=1","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[4107,1,2,1],k=1","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[4096,1,1,1],k=2","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[4107,1,2,1],k=2","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[4096,1,1,1],k=3","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[4107,1,2,1],k=3","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[4096,1,1,1],k=7","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[4107,1,2,1],k=7","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[4096,1,1,1],k=15","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[4107,1,2,1],k=15","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[4096,1,1,1],k=100","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[4107,1,2,1],k=100","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[4096,1,1,1],k=500","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[4107,1,2,1],k=500","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[4096,1,1,1],k=1023","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[4107,1,2,1],k=1023","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[8192,1,1,1],k=1","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[8203,1,2,1],k=1","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[8192,1,1,1],k=2","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[8203,1,2,1],k=2","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[8192,1,1,1],k=3","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[8203,1,2,1],k=3","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[8192,1,1,1],k=7","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[8203,1,2,1],k=7","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[8192,1,1,1],k=15","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[8203,1,2,1],k=15","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[8192,1,1,1],k=100","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[8203,1,2,1],k=100","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[8192,1,1,1],k=500","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[8203,1,2,1],k=500","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[8192,1,1,1],k=1023","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[8203,1,2,1],k=1023","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[16384,1,1,1],k=1","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[16395,1,2,1],k=1","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[16384,1,1,1],k=2","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[16395,1,2,1],k=2","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[16384,1,1,1],k=3","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[16395,1,2,1],k=3","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[16384,1,1,1],k=7","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[16395,1,2,1],k=7","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[16384,1,1,1],k=15","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[16395,1,2,1],k=15","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[16384,1,1,1],k=100","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[16395,1,2,1],k=100","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[16384,1,1,1],k=500","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[16395,1,2,1],k=500","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[16384,1,1,1],k=1023","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[16395,1,2,1],k=1023","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[16384,1,1,1],k=9999","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[16395,1,2,1],k=9999","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[32768,1,1,1],k=1","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[32779,1,2,1],k=1","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[32768,1,1,1],k=2","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[32779,1,2,1],k=2","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[32768,1,1,1],k=3","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[32779,1,2,1],k=3","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[32768,1,1,1],k=7","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[32779,1,2,1],k=7","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[32768,1,1,1],k=15","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[32779,1,2,1],k=15","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[32768,1,1,1],k=100","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[32779,1,2,1],k=100","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[32768,1,1,1],k=500","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[32779,1,2,1],k=500","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[32768,1,1,1],k=1023","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[32779,1,2,1],k=1023","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[32768,1,1,1],k=9999","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[32779,1,2,1],k=9999","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[65536,1,1,1],k=1","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[65547,1,2,1],k=1","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[65536,1,1,1],k=2","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[65547,1,2,1],k=2","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[65536,1,1,1],k=3","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[65547,1,2,1],k=3","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[65536,1,1,1],k=7","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[65547,1,2,1],k=7","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[65536,1,1,1],k=15","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[65547,1,2,1],k=15","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[65536,1,1,1],k=100","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[65547,1,2,1],k=100","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[65536,1,1,1],k=500","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[65547,1,2,1],k=500","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[65536,1,1,1],k=1023","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[65547,1,2,1],k=1023","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[65536,1,1,1],k=9999","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[65547,1,2,1],k=9999","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[131072,1,1,1],k=1","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[131083,1,2,1],k=1","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[131072,1,1,1],k=2","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[131083,1,2,1],k=2","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[131072,1,1,1],k=3","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[131083,1,2,1],k=3","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[131072,1,1,1],k=7","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[131083,1,2,1],k=7","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[131072,1,1,1],k=15","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[131083,1,2,1],k=15","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[131072,1,1,1],k=100","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[131083,1,2,1],k=100","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[131072,1,1,1],k=500","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[131083,1,2,1],k=500","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[131072,1,1,1],k=1023","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[131083,1,2,1],k=1023","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[131072,1,1,1],k=9999","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[131083,1,2,1],k=9999","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[262144,1,1,1],k=1","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[262155,1,2,1],k=1","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[262144,1,1,1],k=2","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[262155,1,2,1],k=2","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[262144,1,1,1],k=3","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[262155,1,2,1],k=3","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[262144,1,1,1],k=7","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[262155,1,2,1],k=7","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[262144,1,1,1],k=15","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[262155,1,2,1],k=15","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[262144,1,1,1],k=100","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[262155,1,2,1],k=100","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[262144,1,1,1],k=500","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[262155,1,2,1],k=500","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[262144,1,1,1],k=1023","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[262155,1,2,1],k=1023","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[262144,1,1,1],k=9999","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[262155,1,2,1],k=9999","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[524288,1,1,1],k=1","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[524299,1,2,1],k=1","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[524288,1,1,1],k=2","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[524299,1,2,1],k=2","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[524288,1,1,1],k=3","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[524299,1,2,1],k=3","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[524288,1,1,1],k=7","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[524299,1,2,1],k=7","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[524288,1,1,1],k=15","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[524299,1,2,1],k=15","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[524288,1,1,1],k=100","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[524299,1,2,1],k=100","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[524288,1,1,1],k=500","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[524299,1,2,1],k=500","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[524288,1,1,1],k=1023","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[524299,1,2,1],k=1023","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[524288,1,1,1],k=9999","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[524299,1,2,1],k=9999","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[16,10,10,10],k=1","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[60,10,10,10],k=1","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[1023,2,1,3],k=1","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[1024,2,1,3],k=1","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[1025,2,1,3],k=1","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[16384,1,1,1],k=1","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[2047,2,1,3],k=1","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[2048,2,1,3],k=1","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[2049,2,1,3],k=1","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[16,10,10,10],k=2","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[60,10,10,10],k=2","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[1023,2,1,3],k=2","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[1024,2,1,3],k=2","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[1025,2,1,3],k=2","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[16384,1,1,1],k=2","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[2047,2,1,3],k=2","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[2048,2,1,3],k=2","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[2049,2,1,3],k=2","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[16,10,10,10],k=3","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[60,10,10,10],k=3","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[1023,2,1,3],k=3","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[1024,2,1,3],k=3","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[1025,2,1,3],k=3","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[16384,1,1,1],k=3","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[2047,2,1,3],k=3","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[2048,2,1,3],k=3","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[2049,2,1,3],k=3","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[16,10,10,10],k=7","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[60,10,10,10],k=7","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[1023,2,1,3],k=7","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[1024,2,1,3],k=7","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[1025,2,1,3],k=7","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[16384,1,1,1],k=7","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[2047,2,1,3],k=7","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[2048,2,1,3],k=7","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[2049,2,1,3],k=7","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[16,10,10,10],k=15","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[60,10,10,10],k=15","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[1023,2,1,3],k=15","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[1024,2,1,3],k=15","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[1025,2,1,3],k=15","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[16384,1,1,1],k=15","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[2047,2,1,3],k=15","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[2048,2,1,3],k=15","support","1","yes","Metal"
-"Metal","TOP_K","type=f32,ne=[2049,2,1,3],k=15","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[1,1,1,1],k=1,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[12,1,2,1],k=1,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[2,1,1,1],k=1,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[13,1,2,1],k=1,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[2,1,1,1],k=2,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[13,1,2,1],k=2,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[4,1,1,1],k=1,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[15,1,2,1],k=1,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[4,1,1,1],k=2,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[15,1,2,1],k=2,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[4,1,1,1],k=3,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[15,1,2,1],k=3,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[8,1,1,1],k=1,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[19,1,2,1],k=1,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[8,1,1,1],k=2,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[19,1,2,1],k=2,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[8,1,1,1],k=3,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[19,1,2,1],k=3,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[8,1,1,1],k=7,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[19,1,2,1],k=7,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[16,1,1,1],k=1,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[27,1,2,1],k=1,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[16,1,1,1],k=2,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[27,1,2,1],k=2,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[16,1,1,1],k=3,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[27,1,2,1],k=3,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[16,1,1,1],k=7,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[27,1,2,1],k=7,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[16,1,1,1],k=15,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[27,1,2,1],k=15,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[32,1,1,1],k=1,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[43,1,2,1],k=1,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[32,1,1,1],k=2,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[43,1,2,1],k=2,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[32,1,1,1],k=3,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[43,1,2,1],k=3,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[32,1,1,1],k=7,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[43,1,2,1],k=7,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[32,1,1,1],k=15,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[43,1,2,1],k=15,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[64,1,1,1],k=1,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[75,1,2,1],k=1,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[64,1,1,1],k=2,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[75,1,2,1],k=2,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[64,1,1,1],k=3,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[75,1,2,1],k=3,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[64,1,1,1],k=7,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[75,1,2,1],k=7,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[64,1,1,1],k=15,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[75,1,2,1],k=15,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[128,1,1,1],k=1,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[139,1,2,1],k=1,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[128,1,1,1],k=2,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[139,1,2,1],k=2,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[128,1,1,1],k=3,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[139,1,2,1],k=3,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[128,1,1,1],k=7,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[139,1,2,1],k=7,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[128,1,1,1],k=15,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[139,1,2,1],k=15,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[128,1,1,1],k=100,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[139,1,2,1],k=100,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[256,1,1,1],k=1,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[267,1,2,1],k=1,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[256,1,1,1],k=2,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[267,1,2,1],k=2,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[256,1,1,1],k=3,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[267,1,2,1],k=3,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[256,1,1,1],k=7,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[267,1,2,1],k=7,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[256,1,1,1],k=15,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[267,1,2,1],k=15,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[256,1,1,1],k=100,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[267,1,2,1],k=100,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[512,1,1,1],k=1,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[523,1,2,1],k=1,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[512,1,1,1],k=2,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[523,1,2,1],k=2,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[512,1,1,1],k=3,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[523,1,2,1],k=3,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[512,1,1,1],k=7,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[523,1,2,1],k=7,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[512,1,1,1],k=15,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[523,1,2,1],k=15,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[512,1,1,1],k=100,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[523,1,2,1],k=100,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[512,1,1,1],k=500,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[523,1,2,1],k=500,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[1024,1,1,1],k=1,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[1035,1,2,1],k=1,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[1024,1,1,1],k=2,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[1035,1,2,1],k=2,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[1024,1,1,1],k=3,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[1035,1,2,1],k=3,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[1024,1,1,1],k=7,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[1035,1,2,1],k=7,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[1024,1,1,1],k=15,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[1035,1,2,1],k=15,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[1024,1,1,1],k=100,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[1035,1,2,1],k=100,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[1024,1,1,1],k=500,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[1035,1,2,1],k=500,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[1024,1,1,1],k=1023,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[1035,1,2,1],k=1023,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[2048,1,1,1],k=1,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[2059,1,2,1],k=1,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[2048,1,1,1],k=2,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[2059,1,2,1],k=2,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[2048,1,1,1],k=3,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[2059,1,2,1],k=3,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[2048,1,1,1],k=7,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[2059,1,2,1],k=7,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[2048,1,1,1],k=15,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[2059,1,2,1],k=15,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[2048,1,1,1],k=100,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[2059,1,2,1],k=100,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[2048,1,1,1],k=500,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[2059,1,2,1],k=500,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[2048,1,1,1],k=1023,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[2059,1,2,1],k=1023,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[4096,1,1,1],k=1,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[4107,1,2,1],k=1,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[4096,1,1,1],k=2,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[4107,1,2,1],k=2,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[4096,1,1,1],k=3,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[4107,1,2,1],k=3,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[4096,1,1,1],k=7,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[4107,1,2,1],k=7,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[4096,1,1,1],k=15,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[4107,1,2,1],k=15,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[4096,1,1,1],k=100,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[4107,1,2,1],k=100,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[4096,1,1,1],k=500,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[4107,1,2,1],k=500,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[4096,1,1,1],k=1023,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[4107,1,2,1],k=1023,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[8192,1,1,1],k=1,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[8203,1,2,1],k=1,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[8192,1,1,1],k=2,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[8203,1,2,1],k=2,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[8192,1,1,1],k=3,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[8203,1,2,1],k=3,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[8192,1,1,1],k=7,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[8203,1,2,1],k=7,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[8192,1,1,1],k=15,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[8203,1,2,1],k=15,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[8192,1,1,1],k=100,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[8203,1,2,1],k=100,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[8192,1,1,1],k=500,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[8203,1,2,1],k=500,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[8192,1,1,1],k=1023,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[8203,1,2,1],k=1023,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[16384,1,1,1],k=1,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[16395,1,2,1],k=1,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[16384,1,1,1],k=2,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[16395,1,2,1],k=2,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[16384,1,1,1],k=3,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[16395,1,2,1],k=3,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[16384,1,1,1],k=7,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[16395,1,2,1],k=7,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[16384,1,1,1],k=15,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[16395,1,2,1],k=15,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[16384,1,1,1],k=100,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[16395,1,2,1],k=100,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[16384,1,1,1],k=500,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[16395,1,2,1],k=500,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[16384,1,1,1],k=1023,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[16395,1,2,1],k=1023,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[16384,1,1,1],k=9999,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[16395,1,2,1],k=9999,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[32768,1,1,1],k=1,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[32779,1,2,1],k=1,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[32768,1,1,1],k=2,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[32779,1,2,1],k=2,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[32768,1,1,1],k=3,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[32779,1,2,1],k=3,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[32768,1,1,1],k=7,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[32779,1,2,1],k=7,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[32768,1,1,1],k=15,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[32779,1,2,1],k=15,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[32768,1,1,1],k=100,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[32779,1,2,1],k=100,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[32768,1,1,1],k=500,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[32779,1,2,1],k=500,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[32768,1,1,1],k=1023,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[32779,1,2,1],k=1023,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[32768,1,1,1],k=9999,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[32779,1,2,1],k=9999,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[65536,1,1,1],k=1,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[65547,1,2,1],k=1,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[65536,1,1,1],k=2,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[65547,1,2,1],k=2,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[65536,1,1,1],k=3,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[65547,1,2,1],k=3,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[65536,1,1,1],k=7,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[65547,1,2,1],k=7,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[65536,1,1,1],k=15,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[65547,1,2,1],k=15,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[65536,1,1,1],k=100,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[65547,1,2,1],k=100,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[65536,1,1,1],k=500,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[65547,1,2,1],k=500,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[65536,1,1,1],k=1023,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[65547,1,2,1],k=1023,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[65536,1,1,1],k=9999,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[65547,1,2,1],k=9999,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[131072,1,1,1],k=1,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[131083,1,2,1],k=1,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[131072,1,1,1],k=2,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[131083,1,2,1],k=2,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[131072,1,1,1],k=3,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[131083,1,2,1],k=3,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[131072,1,1,1],k=7,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[131083,1,2,1],k=7,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[131072,1,1,1],k=15,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[131083,1,2,1],k=15,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[131072,1,1,1],k=100,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[131083,1,2,1],k=100,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[131072,1,1,1],k=500,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[131083,1,2,1],k=500,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[131072,1,1,1],k=1023,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[131083,1,2,1],k=1023,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[131072,1,1,1],k=9999,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[131083,1,2,1],k=9999,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[262144,1,1,1],k=1,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[262155,1,2,1],k=1,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[262144,1,1,1],k=2,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[262155,1,2,1],k=2,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[262144,1,1,1],k=3,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[262155,1,2,1],k=3,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[262144,1,1,1],k=7,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[262155,1,2,1],k=7,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[262144,1,1,1],k=15,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[262155,1,2,1],k=15,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[262144,1,1,1],k=100,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[262155,1,2,1],k=100,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[262144,1,1,1],k=500,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[262155,1,2,1],k=500,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[262144,1,1,1],k=1023,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[262155,1,2,1],k=1023,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[262144,1,1,1],k=9999,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[262155,1,2,1],k=9999,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[524288,1,1,1],k=1,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[524299,1,2,1],k=1,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[524288,1,1,1],k=2,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[524299,1,2,1],k=2,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[524288,1,1,1],k=3,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[524299,1,2,1],k=3,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[524288,1,1,1],k=7,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[524299,1,2,1],k=7,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[524288,1,1,1],k=15,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[524299,1,2,1],k=15,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[524288,1,1,1],k=100,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[524299,1,2,1],k=100,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[524288,1,1,1],k=500,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[524299,1,2,1],k=500,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[524288,1,1,1],k=1023,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[524299,1,2,1],k=1023,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[524288,1,1,1],k=9999,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[524299,1,2,1],k=9999,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[16,10,10,10],k=1,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[60,10,10,10],k=1,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[1023,2,1,3],k=1,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[1024,2,1,3],k=1,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[1025,2,1,3],k=1,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[16384,1,1,1],k=1,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[2047,2,1,3],k=1,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[2048,2,1,3],k=1,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[2049,2,1,3],k=1,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[16,10,10,10],k=2,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[60,10,10,10],k=2,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[1023,2,1,3],k=2,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[1024,2,1,3],k=2,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[1025,2,1,3],k=2,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[16384,1,1,1],k=2,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[2047,2,1,3],k=2,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[2048,2,1,3],k=2,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[2049,2,1,3],k=2,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[16,10,10,10],k=3,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[60,10,10,10],k=3,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[1023,2,1,3],k=3,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[1024,2,1,3],k=3,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[1025,2,1,3],k=3,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[16384,1,1,1],k=3,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[2047,2,1,3],k=3,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[2048,2,1,3],k=3,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[2049,2,1,3],k=3,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[16,10,10,10],k=7,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[60,10,10,10],k=7,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[1023,2,1,3],k=7,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[1024,2,1,3],k=7,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[1025,2,1,3],k=7,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[16384,1,1,1],k=7,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[2047,2,1,3],k=7,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[2048,2,1,3],k=7,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[2049,2,1,3],k=7,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[16,10,10,10],k=15,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[60,10,10,10],k=15,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[1023,2,1,3],k=15,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[1024,2,1,3],k=15,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[1025,2,1,3],k=15,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[16384,1,1,1],k=15,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[2047,2,1,3],k=15,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[2048,2,1,3],k=15,ties=0","support","1","yes","Metal"
+"Metal","TOP_K","type=f32,ne=[2049,2,1,3],k=15,ties=0","support","1","yes","Metal"
 "Metal","UPSCALE","type=f32,ne=[512,512,3,2],scale_factor=2,mode=nearest,transpose=0","support","1","yes","Metal"
 "Metal","UPSCALE","type=f32,ne=[512,512,3,2],scale_factor=2,mode=nearest,transpose=1","support","1","yes","Metal"
 "Metal","UPSCALE","type=f32,ne=[2,5,7,11],ne_tgt=[5,7,11,13],mode=nearest,flags=none","support","1","yes","Metal"
@ -9891,8 +9904,9 @@
 "Metal","GROUP_NORM","type=f32,ne=[64,64,320,1],num_groups=32,eps=0.000001","support","1","yes","Metal"
 "Metal","GROUP_NORM","type=f32,ne=[9,9,1280,1],num_groups=32,eps=0.000001","support","1","yes","Metal"
 "Metal","ACC","type=f32,ne_a=[256,17,1,1],ne_b=[256,16,1,1]","support","1","yes","Metal"
-"Metal","PAD","type=f32,ne_a=[512,512,1,1],pad_0=1,pad_1=1","support","1","yes","Metal"
-"Metal","PAD","type=f32,ne_a=[512,512,3,1],lp0=1,rp0=1,lp1=1,rp1=1,lp2=1,rp2=1,lp3=1,rp3=1,v=0","support","0","no","Metal"
+"Metal","PAD","type=f32,ne_a=[512,512,1,1],pad_0=1,pad_1=1,circular=0","support","1","yes","Metal"
+"Metal","PAD","type=f32,ne_a=[33,17,2,1],pad_0=4,pad_1=3,circular=1","support","0","no","Metal"
+"Metal","PAD","type=f32,ne_a=[512,512,3,1],lp0=1,rp0=1,lp1=1,rp1=1,lp2=1,rp2=1,lp3=1,rp3=1,v=0,circular=0","support","0","no","Metal"
 "Metal","PAD_REFLECT_1D","type=f32,ne_a=[512,34,2,1],pad_0=10,pad_1=9","support","1","yes","Metal"
 "Metal","PAD_REFLECT_1D","type=f32,ne_a=[3000,384,4,1],pad_0=10,pad_1=9","support","1","yes","Metal"
 "Metal","ROLL","shift0=3,shift1=-2,shift3=1,shift4=-1","support","0","no","Metal"
@ -9923,17 +9937,41 @@
 "Metal","FILL","type=f32,ne=[303,207,11,3],c=2.000000","support","1","yes","Metal"
 "Metal","FILL","type=f32,ne=[800,600,4,4],c=-152.000000","support","1","yes","Metal"
 "Metal","FILL","type=f32,ne=[2048,512,2,2],c=3.500000","support","1","yes","Metal"
+"Metal","DIAG","type=f32,ne=[10,1,4,3]","support","0","no","Metal"
+"Metal","DIAG","type=f32,ne=[79,1,19,13]","support","0","no","Metal"
+"Metal","DIAG","type=f32,ne=[256,1,8,16]","support","0","no","Metal"
 "Metal","SOLVE_TRI","type=f32,ne_lhs=[10,10,4,3],ne_rhs=[3,10,4,3]","support","0","no","Metal"
 "Metal","SOLVE_TRI","type=f32,ne_lhs=[11,11,1,1],ne_rhs=[5,11,1,1]","support","0","no","Metal"
 "Metal","SOLVE_TRI","type=f32,ne_lhs=[17,17,2,4],ne_rhs=[9,17,2,4]","support","0","no","Metal"
 "Metal","SOLVE_TRI","type=f32,ne_lhs=[30,30,7,1],ne_rhs=[8,30,7,1]","support","0","no","Metal"
 "Metal","SOLVE_TRI","type=f32,ne_lhs=[42,42,5,2],ne_rhs=[10,42,5,2]","support","0","no","Metal"
 "Metal","SOLVE_TRI","type=f32,ne_lhs=[64,64,2,2],ne_rhs=[10,64,2,2]","support","0","no","Metal"
+"Metal","SOLVE_TRI","type=f32,ne_lhs=[64,64,2,2],ne_rhs=[64,64,2,2]","support","0","no","Metal"
+"Metal","SOLVE_TRI","type=f32,ne_lhs=[79,79,5,3],ne_rhs=[417,79,5,3]","support","0","no","Metal"
+"Metal","SOLVE_TRI","type=f32,ne_lhs=[128,128,4,2],ne_rhs=[32,128,4,2]","support","0","no","Metal"
+"Metal","SOLVE_TRI","type=f32,ne_lhs=[80,80,2,8],ne_rhs=[80,80,2,8]","support","0","no","Metal"
+"Metal","SOLVE_TRI","type=f32,ne_lhs=[80,80,2,8],ne_rhs=[79,80,2,8]","support","0","no","Metal"
+"Metal","SOLVE_TRI","type=f32,ne_lhs=[80,80,2,8],ne_rhs=[81,80,2,8]","support","0","no","Metal"
+"Metal","SOLVE_TRI","type=f32,ne_lhs=[80,80,8,8],ne_rhs=[80,80,8,8]","support","0","no","Metal"
+"Metal","SOLVE_TRI","type=f32,ne_lhs=[80,80,8,8],ne_rhs=[79,80,8,8]","support","0","no","Metal"
+"Metal","SOLVE_TRI","type=f32,ne_lhs=[80,80,8,8],ne_rhs=[81,80,8,8]","support","0","no","Metal"
+"Metal","SOLVE_TRI","type=f32,ne_lhs=[84,84,4,4],ne_rhs=[32,84,4,4]","support","0","no","Metal"
+"Metal","SOLVE_TRI","type=f32,ne_lhs=[95,95,8,8],ne_rhs=[40,95,8,8]","support","0","no","Metal"
 "Metal","SOLVE_TRI","type=f32,ne_lhs=[100,100,4,4],ne_rhs=[41,100,4,4]","support","0","no","Metal"
-"Metal","PAD","type=f32,ne_a=[512,512,1,1],lp0=0,rp0=1,lp1=0,rp1=1,lp2=0,rp2=0,lp3=0,rp3=0,v=0","support","1","yes","Metal"
-"Metal","PAD","type=f32,ne_a=[11,22,33,44],lp0=1,rp0=2,lp1=3,rp1=4,lp2=5,rp2=6,lp3=7,rp3=8,v=0","support","0","no","Metal"
-"Metal","PAD","type=f32,ne_a=[512,512,1,1],lp0=0,rp0=1,lp1=0,rp1=1,lp2=0,rp2=0,lp3=0,rp3=0,v=1","support","1","yes","Metal"
-"Metal","PAD","type=f32,ne_a=[11,22,33,44],lp0=1,rp0=2,lp1=3,rp1=4,lp2=5,rp2=6,lp3=7,rp3=8,v=1","support","0","no","Metal"
+"Metal","SOLVE_TRI","type=f32,ne_lhs=[128,128,4,4],ne_rhs=[31,128,4,4]","support","0","no","Metal"
+"Metal","SOLVE_TRI","type=f32,ne_lhs=[128,128,4,4],ne_rhs=[32,128,4,4]","support","0","no","Metal"
+"Metal","SOLVE_TRI","type=f32,ne_lhs=[128,128,3,4],ne_rhs=[32,128,3,4]","support","0","no","Metal"
+"Metal","SOLVE_TRI","type=f32,ne_lhs=[128,128,4,1],ne_rhs=[32,128,4,1]","support","0","no","Metal"
+"Metal","SOLVE_TRI","type=f32,ne_lhs=[64,64,4,4],ne_rhs=[200,64,4,4]","support","0","no","Metal"
+"Metal","SOLVE_TRI","type=f32,ne_lhs=[64,64,4,4],ne_rhs=[384,64,4,4]","support","0","no","Metal"
+"Metal","PAD","type=f32,ne_a=[512,512,1,1],lp0=0,rp0=1,lp1=0,rp1=1,lp2=0,rp2=0,lp3=0,rp3=0,v=0,circular=0","support","1","yes","Metal"
+"Metal","PAD","type=f32,ne_a=[11,22,33,44],lp0=1,rp0=2,lp1=3,rp1=4,lp2=5,rp2=6,lp3=7,rp3=8,v=0,circular=0","support","0","no","Metal"
+"Metal","PAD","type=f32,ne_a=[512,512,1,1],lp0=0,rp0=1,lp1=0,rp1=1,lp2=0,rp2=0,lp3=0,rp3=0,v=0,circular=1","support","0","no","Metal"
+"Metal","PAD","type=f32,ne_a=[11,22,33,44],lp0=1,rp0=2,lp1=3,rp1=4,lp2=5,rp2=6,lp3=7,rp3=8,v=0,circular=1","support","0","no","Metal"
+"Metal","PAD","type=f32,ne_a=[512,512,1,1],lp0=0,rp0=1,lp1=0,rp1=1,lp2=0,rp2=0,lp3=0,rp3=0,v=1,circular=0","support","1","yes","Metal"
+"Metal","PAD","type=f32,ne_a=[11,22,33,44],lp0=1,rp0=2,lp1=3,rp1=4,lp2=5,rp2=6,lp3=7,rp3=8,v=1,circular=0","support","0","no","Metal"
+"Metal","PAD","type=f32,ne_a=[512,512,1,1],lp0=0,rp0=1,lp1=0,rp1=1,lp2=0,rp2=0,lp3=0,rp3=0,v=1,circular=1","support","0","no","Metal"
+"Metal","PAD","type=f32,ne_a=[11,22,33,44],lp0=1,rp0=2,lp1=3,rp1=4,lp2=5,rp2=6,lp3=7,rp3=8,v=1,circular=1","support","0","no","Metal"
 "Metal","FLASH_ATTN_EXT","hsk=40,hsv=40,nh=4,nr23=[1,1],kv=113,nb=1,mask=1,sinks=1,max_bias=0.000000,logit_softcap=0.000000,prec=f32,type_KV=f32,permute=[0,1,2,3]","support","1","yes","Metal"
 "Metal","FLASH_ATTN_EXT","hsk=40,hsv=40,nh=4,nr23=[1,1],kv=113,nb=1,mask=1,sinks=1,max_bias=0.000000,logit_softcap=0.000000,prec=f32,type_KV=f16,permute=[0,1,2,3]","support","1","yes","Metal"
 "Metal","FLASH_ATTN_EXT","hsk=40,hsv=40,nh=4,nr23=[1,1],kv=113,nb=1,mask=1,sinks=1,max_bias=0.000000,logit_softcap=0.000000,prec=f32,type_KV=bf16,permute=[0,1,2,3]","support","1","yes","Metal"
--- a/ggml/CMakeLists.txt
+++ b/ggml/CMakeLists.txt
@ -4,7 +4,7 @@ project("ggml" C CXX ASM)
 ### GGML Version
 set(GGML_VERSION_MAJOR 0)
 set(GGML_VERSION_MINOR 9)
-set(GGML_VERSION_PATCH 4)
+set(GGML_VERSION_PATCH 5)
 set(GGML_VERSION_BASE "${GGML_VERSION_MAJOR}.${GGML_VERSION_MINOR}.${GGML_VERSION_PATCH}")

 find_program(GIT_EXE NAMES git git.exe NO_CMAKE_FIND_ROOT_PATH)
--- a/ggml/include/ggml-backend.h
+++ b/ggml/include/ggml-backend.h
@ -358,7 +358,7 @@ extern "C" {
    typedef bool (*ggml_backend_eval_callback)(int node_index, struct ggml_tensor * t1, struct ggml_tensor * t2, void * user_data);

    // Compare the output of two backends
-    GGML_API bool ggml_backend_compare_graph_backend(ggml_backend_t backend1, ggml_backend_t backend2, struct ggml_cgraph * graph, ggml_backend_eval_callback callback, void * user_data, struct ggml_tensor * test_node);
+    GGML_API bool ggml_backend_compare_graph_backend(ggml_backend_t backend1, ggml_backend_t backend2, struct ggml_cgraph * graph, ggml_backend_eval_callback callback, void * user_data, struct ggml_tensor const * const * test_nodes, size_t num_test_nodes);

    // Tensor initialization
    GGML_API enum ggml_status ggml_backend_tensor_alloc(ggml_backend_buffer_t buffer, struct ggml_tensor * tensor, void * addr);
--- a/ggml/src/ggml-backend.cpp
+++ b/ggml/src/ggml-backend.cpp
@ -2053,7 +2053,7 @@ void ggml_backend_graph_copy_free(struct ggml_backend_graph_copy copy) {
    ggml_free(copy.ctx_unallocated);
 }

-bool ggml_backend_compare_graph_backend(ggml_backend_t backend1, ggml_backend_t backend2, struct ggml_cgraph * graph, ggml_backend_eval_callback callback, void * user_data, struct ggml_tensor * test_node) {
+bool ggml_backend_compare_graph_backend(ggml_backend_t backend1, ggml_backend_t backend2, struct ggml_cgraph * graph, ggml_backend_eval_callback callback, void * user_data, struct ggml_tensor const * const * test_nodes, size_t num_test_nodes) {
    struct ggml_backend_graph_copy copy = ggml_backend_graph_copy(backend2, graph);
    if (copy.buffer == NULL) {
        return false;
@ -2064,22 +2064,22 @@ bool ggml_backend_compare_graph_backend(ggml_backend_t backend1, ggml_backend_t

    assert(g1->n_nodes == g2->n_nodes);

-    if (test_node != nullptr) {
-        // Compute the whole graph and only test the output for a specific tensor
+    if (num_test_nodes != 0) {
+        GGML_ASSERT(test_nodes);
+        // Compute the whole graph and only test the output for specific tensors
        ggml_backend_graph_compute(backend1, g1);
        ggml_backend_graph_compute(backend2, g2);

-        int test_node_idx = -1;
+        bool verified = false;
        for (int i = 0; i < g1->n_nodes; i++) {
-            struct ggml_tensor * t1 = g1->nodes[i];
-            if (t1 == test_node) {
-                test_node_idx = i;
-                break;
+            for (size_t j = 0; j < num_test_nodes; ++j) {
+                if (g1->nodes[i] == test_nodes[j]) {
+                    callback(i, g1->nodes[i], g2->nodes[i], user_data);
+                    verified = true;
+                }
            }
        }
-        GGML_ASSERT(test_node_idx != -1);
-
-        callback(test_node_idx, g1->nodes[test_node_idx], g2->nodes[test_node_idx], user_data);
+        GGML_ASSERT(verified);
    } else {
        for (int i = 0; i < g1->n_nodes; i++) {
            struct ggml_tensor * t1 = g1->nodes[i];
--- a/ggml/src/ggml-cuda/fattn-mma-f16.cuh
+++ b/ggml/src/ggml-cuda/fattn-mma-f16.cuh
@ -531,7 +531,7 @@ static __device__ __forceinline__ void flash_attn_ext_f16_iter(
        for (int k0 = 0; k0 < nbatch_fa; k0 += np*T_C_KQ::I) {
 #pragma unroll
            for (int l = 0; l < T_C_KQ::ne; ++l) {
-                if (!oob_check || k0 + T_C_KQ::get_i(l) < k_VKQ_sup) {
+                if (!oob_check || k0 + (threadIdx.y % np)*T_C_KQ::I + T_C_KQ::get_i(l) < k_VKQ_sup) {
                    KQ_max_new[l % 2] = fmaxf(KQ_max_new[l % 2], KQ_C[k0/(np*T_C_KQ::I)].x[l] + FATTN_KQ_MAX_OFFSET);
                }
            }
@ -583,7 +583,7 @@ static __device__ __forceinline__ void flash_attn_ext_f16_iter(
        for (int k0 = 0; k0 < nbatch_fa; k0 += np*T_C_KQ::J) {
 #pragma unroll
            for (int l = 0; l < T_C_KQ::ne; ++l) {
-                if (!oob_check || k0 + T_C_KQ::get_j(l) < k_VKQ_sup) {
+                if (!oob_check || k0 + (threadIdx.y % np)*T_C_KQ::J + T_C_KQ::get_j(l) < k_VKQ_sup) {
                    // Turing + Volta:
                    KQ_max_new[(l/2) % 2] = fmaxf(KQ_max_new[(l/2) % 2], KQ_C[(k0/(np*T_C_KQ::J))].x[l] + FATTN_KQ_MAX_OFFSET);
                }
--- a/ggml/src/ggml-cuda/ggml-cuda.cu
+++ b/ggml/src/ggml-cuda/ggml-cuda.cu
@ -201,16 +201,6 @@ static ggml_cuda_device_info ggml_cuda_init() {
    GGML_ASSERT(info.device_count <= GGML_CUDA_MAX_DEVICES);

    int64_t total_vram = 0;
-#ifdef GGML_CUDA_FORCE_MMQ
-    GGML_LOG_INFO("%s: GGML_CUDA_FORCE_MMQ:    yes\n", __func__);
-#else
-    GGML_LOG_INFO("%s: GGML_CUDA_FORCE_MMQ:    no\n", __func__);
-#endif // GGML_CUDA_FORCE_MMQ
-#ifdef GGML_CUDA_FORCE_CUBLAS
-    GGML_LOG_INFO("%s: GGML_CUDA_FORCE_CUBLAS: yes\n", __func__);
-#else
-    GGML_LOG_INFO("%s: GGML_CUDA_FORCE_CUBLAS: no\n", __func__);
-#endif // GGML_CUDA_FORCE_CUBLAS
    GGML_LOG_INFO("%s: found %d " GGML_CUDA_NAME " devices:\n", __func__, info.device_count);

    std::vector<std::pair<int, std::string>> turing_devices_without_mma;
--- a/ggml/src/ggml-metal/ggml-metal-device.cpp
+++ b/ggml/src/ggml-metal/ggml-metal-device.cpp
@ -1684,3 +1684,60 @@ ggml_metal_pipeline_with_params ggml_metal_library_get_pipeline_opt_step_sgd(ggm

    return res;
 }
+
+ggml_metal_pipeline_with_params ggml_metal_library_get_pipeline_memset(ggml_metal_library_t lib, const ggml_tensor *  op) {
+    GGML_ASSERT(op->type == GGML_TYPE_I64);
+
+    char base[256];
+    char name[256];
+
+    snprintf(base, 256, "kernel_memset_%s", ggml_type_name(op->type));
+    snprintf(name, 256, "%s", base);
+
+    ggml_metal_pipeline_with_params res = ggml_metal_library_get_pipeline(lib, name);
+    if (!res.pipeline) {
+        res = ggml_metal_library_compile_pipeline(lib, base, name, nullptr);
+    }
+
+    return res;
+}
+
+ggml_metal_pipeline_with_params ggml_metal_library_get_pipeline_count_equal(ggml_metal_library_t lib, const ggml_tensor *  op) {
+    assert(op->op == GGML_OP_COUNT_EQUAL);
+
+    GGML_TENSOR_LOCALS(int64_t, ne0, op->src[0], ne);
+
+    GGML_ASSERT(op->src[0]->type == op->src[1]->type);
+    GGML_ASSERT(op->src[0]->type == GGML_TYPE_I32);
+    GGML_ASSERT(op->type == GGML_TYPE_I64);
+
+    // note: the kernel only supports i32 output due to metal atomic add only supporting atomic_int
+    GGML_ASSERT(ggml_nelements(op->src[0]) < (1LL << 31));
+
+    char base[256];
+    char name[256];
+
+    int nsg = 1;
+    while (32*nsg < ne00 && nsg < 32) {
+        nsg *= 2;
+    }
+
+    snprintf(base, 256, "kernel_count_equal_%s", ggml_type_name(op->src[0]->type));
+    snprintf(name, 256, "%s_nsg=%d", base, nsg);
+
+    ggml_metal_pipeline_with_params res = ggml_metal_library_get_pipeline(lib, name);
+    if (!res.pipeline) {
+        ggml_metal_cv_t cv = ggml_metal_cv_init();
+
+        ggml_metal_cv_set_int16(cv, nsg, FC_COUNT_EQUAL + 0);
+
+        res = ggml_metal_library_compile_pipeline(lib, base, name, cv);
+
+        ggml_metal_cv_free(cv);
+    }
+
+    res.smem = 32 * sizeof(int32_t);
+    res.nsg  = nsg;
+
+    return res;
+}
--- a/ggml/src/ggml-metal/ggml-metal-device.h
+++ b/ggml/src/ggml-metal/ggml-metal-device.h
@ -147,6 +147,8 @@ struct ggml_metal_pipeline_with_params ggml_metal_library_get_pipeline_arange
 struct ggml_metal_pipeline_with_params ggml_metal_library_get_pipeline_timestep_embedding(ggml_metal_library_t lib, const struct ggml_tensor * op);
 struct ggml_metal_pipeline_with_params ggml_metal_library_get_pipeline_opt_step_adamw    (ggml_metal_library_t lib, const struct ggml_tensor * op);
 struct ggml_metal_pipeline_with_params ggml_metal_library_get_pipeline_opt_step_sgd      (ggml_metal_library_t lib, const struct ggml_tensor * op);
+struct ggml_metal_pipeline_with_params ggml_metal_library_get_pipeline_memset            (ggml_metal_library_t lib, const struct ggml_tensor * op);
+struct ggml_metal_pipeline_with_params ggml_metal_library_get_pipeline_count_equal       (ggml_metal_library_t lib, const struct ggml_tensor * op);

 struct ggml_metal_pipeline_with_params ggml_metal_library_get_pipeline_flash_attn_ext_pad(
        ggml_metal_library_t lib,
--- a/ggml/src/ggml-metal/ggml-metal-device.m
+++ b/ggml/src/ggml-metal/ggml-metal-device.m
@ -1023,6 +1023,11 @@ bool ggml_metal_device_supports_op(ggml_metal_device_t dev, const struct ggml_te
            return has_simdgroup_reduction && ggml_is_contiguous_rows(op->src[0]);
        case GGML_OP_L2_NORM:
            return has_simdgroup_reduction && (op->ne[0] % 4 == 0 && ggml_is_contiguous_1(op->src[0]));
+        case GGML_OP_COUNT_EQUAL:
+            return has_simdgroup_reduction &&
+                op->src[0]->type == GGML_TYPE_I32 &&
+                op->src[1]->type == GGML_TYPE_I32 &&
+                op->type == GGML_TYPE_I64;
        case GGML_OP_ARGMAX:
            return has_simdgroup_reduction;
        case GGML_OP_NORM:
--- a/ggml/src/ggml-metal/ggml-metal-impl.h
+++ b/ggml/src/ggml-metal/ggml-metal-impl.h
@ -78,6 +78,7 @@
 #define FC_MUL_MM                      700
 #define FC_ROPE                        800
 #define FC_SSM_CONV                    900
+#define FC_COUNT_EQUAL                 1000

 // op-specific constants
 #define OP_FLASH_ATTN_EXT_NQPTG 8
@ -894,6 +895,25 @@ typedef struct {
    float    step;
 } ggml_metal_kargs_arange;

+typedef struct {
+    int64_t val;
+} ggml_metal_kargs_memset;
+
+typedef struct {
+    int32_t  ne00;
+    int32_t  ne01;
+    int32_t  ne02;
+    int32_t  ne03;
+    uint64_t nb00;
+    uint64_t nb01;
+    uint64_t nb02;
+    uint64_t nb03;
+    uint64_t nb10;
+    uint64_t nb11;
+    uint64_t nb12;
+    uint64_t nb13;
+} ggml_metal_kargs_count_equal;
+
 typedef struct {
    int32_t  k0;
    int32_t  k1;
--- a/ggml/src/ggml-metal/ggml-metal-ops.cpp
+++ b/ggml/src/ggml-metal/ggml-metal-ops.cpp
@ -448,7 +448,11 @@ static int ggml_metal_op_encode_impl(ggml_metal_op_t ctx, int idx) {
            {
                n_fuse = ggml_metal_op_opt_step_sgd(ctx, idx);
            } break;
-       default:
+        case GGML_OP_COUNT_EQUAL:
+            {
+                n_fuse = ggml_metal_op_count_equal(ctx, idx);
+            } break;
+        default:
            {
                GGML_LOG_ERROR("%s: error: node %3d, op = %8s not implemented\n", __func__, idx, ggml_op_name(node->op));
                GGML_ABORT("fatal error");
@ -4090,3 +4094,64 @@ int ggml_metal_op_opt_step_sgd(ggml_metal_op_t ctx, int idx) {

    return 1;
 }
+
+int ggml_metal_op_count_equal(ggml_metal_op_t ctx, int idx) {
+    ggml_tensor * op = ctx->node(idx);
+
+    ggml_metal_library_t lib = ctx->lib;
+    ggml_metal_encoder_t enc = ctx->enc;
+
+    GGML_TENSOR_LOCALS(int32_t,  ne0, op->src[0], ne);
+    GGML_TENSOR_LOCALS(uint64_t, nb0, op->src[0], nb);
+    GGML_TENSOR_LOCALS(uint64_t, nb1, op->src[1], nb);
+
+    {
+        ggml_metal_kargs_memset args = { /*.val =*/ 0 };
+
+        auto pipeline = ggml_metal_library_get_pipeline_memset(lib, op);
+
+        ggml_metal_encoder_set_pipeline(enc, pipeline);
+        ggml_metal_encoder_set_bytes(enc, &args, sizeof(args), 0);
+        ggml_metal_encoder_set_buffer(enc, ggml_metal_get_buffer_id(op), 1);
+
+        ggml_metal_encoder_dispatch_threadgroups(enc, 1, 1, 1, 1, 1, 1);
+    }
+
+    ggml_metal_op_concurrency_reset(ctx);
+
+    {
+        ggml_metal_kargs_count_equal args = {
+            /*.ne00 =*/ ne00,
+            /*.ne01 =*/ ne01,
+            /*.ne02 =*/ ne02,
+            /*.ne03 =*/ ne03,
+            /*.nb00 =*/ nb00,
+            /*.nb01 =*/ nb01,
+            /*.nb02 =*/ nb02,
+            /*.nb03 =*/ nb03,
+            /*.nb10 =*/ nb10,
+            /*.nb11 =*/ nb11,
+            /*.nb12 =*/ nb12,
+            /*.nb13 =*/ nb13,
+        };
+
+        auto pipeline = ggml_metal_library_get_pipeline_count_equal(lib, op);
+
+        const size_t smem = pipeline.smem;
+
+        const int nth = 32*pipeline.nsg;
+
+        GGML_ASSERT(nth <= ggml_metal_pipeline_max_theads_per_threadgroup(pipeline));
+
+        ggml_metal_encoder_set_pipeline(enc, pipeline);
+        ggml_metal_encoder_set_bytes(enc, &args, sizeof(args), 0);
+        ggml_metal_encoder_set_buffer(enc, ggml_metal_get_buffer_id(op->src[0]), 1);
+        ggml_metal_encoder_set_buffer(enc, ggml_metal_get_buffer_id(op->src[1]), 2);
+        ggml_metal_encoder_set_buffer(enc, ggml_metal_get_buffer_id(op), 3);
+
+        ggml_metal_encoder_set_threadgroup_memory_size(enc, smem, 0);
+        ggml_metal_encoder_dispatch_threadgroups(enc, ne01, ne02, ne03, nth, 1, 1);
+    }
+
+    return 1;
+}
--- a/ggml/src/ggml-metal/ggml-metal-ops.h
+++ b/ggml/src/ggml-metal/ggml-metal-ops.h
@ -87,6 +87,7 @@ int ggml_metal_op_leaky_relu        (ggml_metal_op_t ctx, int idx);
 int ggml_metal_op_tri               (ggml_metal_op_t ctx, int idx);
 int ggml_metal_op_opt_step_adamw    (ggml_metal_op_t ctx, int idx);
 int ggml_metal_op_opt_step_sgd      (ggml_metal_op_t ctx, int idx);
+int ggml_metal_op_count_equal       (ggml_metal_op_t ctx, int idx);

 #ifdef __cplusplus
 }
--- a/ggml/src/ggml-metal/ggml-metal.metal
+++ b/ggml/src/ggml-metal/ggml-metal.metal
@ -1790,6 +1790,7 @@ kernel void kernel_op_sum_f32(
        return;
    }

+    // TODO: become function constant
    const uint nsg = (ntg.x + 31) / 32;

    float sumf = 0;
@ -9557,9 +9558,6 @@ template [[host_name("kernel_mul_mm_iq4_xs_f32")]]  kernel mul_mm_t kernel_mul_m

 template [[host_name("kernel_mul_mm_f32_f16")]]     kernel mul_mm_t kernel_mul_mm<half,   half4x4,   simdgroup_half8x8,   half,   half2x4,   simdgroup_half8x8,   float4x4,      1,     dequantize_f32,     float,  float4x4,  half, half2x4>;
 template [[host_name("kernel_mul_mm_f16_f16")]]     kernel mul_mm_t kernel_mul_mm<half,   half4x4,   simdgroup_half8x8,   half,   half2x4,   simdgroup_half8x8,   half4x4,       1,     dequantize_f16,     half,   half4x4,   half, half2x4>;
-#if defined(GGML_METAL_HAS_BF16)
-template [[host_name("kernel_mul_mm_bf16_f16")]]    kernel mul_mm_t kernel_mul_mm<bfloat, bfloat4x4, simdgroup_bfloat8x8, half,   half2x4,   simdgroup_half8x8,   bfloat4x4,     1,     dequantize_bf16,    bfloat, bfloat4x4, half, half2x4>;
-#endif
 template [[host_name("kernel_mul_mm_q4_0_f16")]]    kernel mul_mm_t kernel_mul_mm<half,   half4x4,   simdgroup_half8x8,   half,   half2x4,   simdgroup_half8x8,   block_q4_0,    2,     dequantize_q4_0,    float,  float4x4,  half, half2x4>;
 template [[host_name("kernel_mul_mm_q4_1_f16")]]    kernel mul_mm_t kernel_mul_mm<half,   half4x4,   simdgroup_half8x8,   half,   half2x4,   simdgroup_half8x8,   block_q4_1,    2,     dequantize_q4_1,    float,  float4x4,  half, half2x4>;
 template [[host_name("kernel_mul_mm_q5_0_f16")]]    kernel mul_mm_t kernel_mul_mm<half,   half4x4,   simdgroup_half8x8,   half,   half2x4,   simdgroup_half8x8,   block_q5_0,    2,     dequantize_q5_0,    float,  float4x4,  half, half2x4>;
@ -9615,9 +9613,6 @@ template [[host_name("kernel_mul_mm_id_iq4_xs_f32")]]  kernel mul_mm_id kernel_m

 template [[host_name("kernel_mul_mm_id_f32_f16")]]     kernel mul_mm_id kernel_mul_mm_id<half,   half4x4,   simdgroup_half8x8,   half,   half2x4,   simdgroup_half8x8,   float4x4,      1,     dequantize_f32,     float,  float4x4,  half, half2x4>;
 template [[host_name("kernel_mul_mm_id_f16_f16")]]     kernel mul_mm_id kernel_mul_mm_id<half,   half4x4,   simdgroup_half8x8,   half,   half2x4,   simdgroup_half8x8,   half4x4,       1,     dequantize_f16,     half,   half4x4,   half, half2x4>;
-#if defined(GGML_METAL_HAS_BF16)
-template [[host_name("kernel_mul_mm_id_bf16_f16")]]    kernel mul_mm_id kernel_mul_mm_id<bfloat, bfloat4x4, simdgroup_bfloat8x8, half,   half2x4,   simdgroup_half8x8,   bfloat4x4,     1,     dequantize_bf16,    bfloat, bfloat4x4, half, half2x4>;
-#endif
 template [[host_name("kernel_mul_mm_id_q4_0_f16")]]    kernel mul_mm_id kernel_mul_mm_id<half,   half4x4,   simdgroup_half8x8,   half,   half2x4,   simdgroup_half8x8,   block_q4_0,    2,     dequantize_q4_0,    float,  float4x4,  half, half2x4>;
 template [[host_name("kernel_mul_mm_id_q4_1_f16")]]    kernel mul_mm_id kernel_mul_mm_id<half,   half4x4,   simdgroup_half8x8,   half,   half2x4,   simdgroup_half8x8,   block_q4_1,    2,     dequantize_q4_1,    float,  float4x4,  half, half2x4>;
 template [[host_name("kernel_mul_mm_id_q5_0_f16")]]    kernel mul_mm_id kernel_mul_mm_id<half,   half4x4,   simdgroup_half8x8,   half,   half2x4,   simdgroup_half8x8,   block_q5_0,    2,     dequantize_q5_0,    float,  float4x4,  half, half2x4>;
@ -9920,3 +9915,75 @@ kernel void kernel_opt_step_sgd_f32(

    x[gid] = x[gid] * (1.0f - pars[0] * pars[1]) - pars[0] * g[gid];
 }
+
+template<typename T>
+kernel void kernel_memset(
+        constant ggml_metal_kargs_fill & args,
+        device T * dst,
+        uint tpig[[thread_position_in_grid]]) {
+    dst[tpig] = args.val;
+}
+
+typedef decltype(kernel_memset<int64_t>) kernel_memset_t;
+
+template [[host_name("kernel_memset_i64")]] kernel kernel_memset_t kernel_memset<int64_t>;
+
+constant short FC_count_equal_nsg [[function_constant(FC_COUNT_EQUAL + 0)]];
+
+template<typename T>
+kernel void kernel_count_equal(
+        constant ggml_metal_kargs_count_equal & args,
+        device   const char * src0,
+        device   const char * src1,
+        device   atomic_int * dst,
+        threadgroup int32_t * shmem_i32 [[threadgroup(0)]],
+        uint3   tgpig[[threadgroup_position_in_grid]],
+        ushort3 tpitg[[thread_position_in_threadgroup]],
+        ushort  sgitg[[simdgroup_index_in_threadgroup]],
+        ushort  tiisg[[thread_index_in_simdgroup]],
+        ushort3   ntg[[threads_per_threadgroup]]) {
+    const short NSG = FC_count_equal_nsg;
+
+    const int i3 = tgpig.z;
+    const int i2 = tgpig.y;
+    const int i1 = tgpig.x;
+
+    if (i3 >= args.ne03 || i2 >= args.ne02 || i1 >= args.ne01) {
+        return;
+    }
+
+    int sum = 0;
+
+    device const char * base0 = src0 + i1*args.nb01 + i2*args.nb02 + i3*args.nb03;
+    device const char * base1 = src1 + i1*args.nb11 + i2*args.nb12 + i3*args.nb13;
+
+    for (int64_t i0 = tpitg.x; i0 < args.ne00; i0 += ntg.x) {
+        const T v0 = *(device const T *)(base0 + i0*args.nb00);
+        const T v1 = *(device const T *)(base1 + i0*args.nb10);
+        sum += (v0 == v1);
+    }
+
+    sum = simd_sum(sum);
+
+    if (tiisg == 0) {
+        shmem_i32[sgitg] = sum;
+    }
+
+    threadgroup_barrier(mem_flags::mem_threadgroup);
+
+    if (sgitg == 0) {
+        float v = 0.0f;
+        if (tpitg.x < NSG) {
+            v = shmem_i32[tpitg.x];
+        }
+
+        float total = simd_sum(v);
+        if (tpitg.x == 0) {
+            atomic_fetch_add_explicit(dst, (int32_t) total, memory_order_relaxed);
+        }
+    }
+}
+
+typedef decltype(kernel_count_equal<int32_t>) kernel_count_equal_t;
+
+template [[host_name("kernel_count_equal_i32")]] kernel kernel_count_equal_t kernel_count_equal<int32_t>;
--- a/ggml/src/ggml-vulkan/ggml-vulkan.cpp
+++ b/ggml/src/ggml-vulkan/ggml-vulkan.cpp
@ -434,8 +434,15 @@ static constexpr std::initializer_list<ggml_op> topk_moe_early_softmax_norm{ GGM
                                                                             GGML_OP_VIEW,     GGML_OP_GET_ROWS, GGML_OP_RESHAPE,
                                                                             GGML_OP_SUM_ROWS, GGML_OP_CLAMP,    GGML_OP_DIV,
                                                                             GGML_OP_RESHAPE };
+
+static constexpr std::initializer_list<ggml_op> topk_moe_sigmoid_norm_bias{ GGML_OP_UNARY,    GGML_OP_RESHAPE,  GGML_OP_ADD,
+                                                                            GGML_OP_ARGSORT,  GGML_OP_VIEW,     GGML_OP_GET_ROWS,
+                                                                            GGML_OP_RESHAPE,  GGML_OP_SUM_ROWS, GGML_OP_CLAMP,
+                                                                            GGML_OP_DIV,      GGML_OP_RESHAPE };
+
 static constexpr std::initializer_list<ggml_op> topk_moe_early_softmax     { GGML_OP_SOFT_MAX, GGML_OP_RESHAPE,  GGML_OP_ARGSORT,
                                                                             GGML_OP_VIEW,     GGML_OP_GET_ROWS };
+
 static constexpr std::initializer_list<ggml_op> topk_moe_late_softmax      { GGML_OP_ARGSORT,  GGML_OP_VIEW,
                                                                             GGML_OP_GET_ROWS, GGML_OP_RESHAPE,
                                                                             GGML_OP_SOFT_MAX, GGML_OP_RESHAPE };
@ -464,6 +471,32 @@ static constexpr std::initializer_list<std::array<int, 3>> topk_moe_early_softma
    { 9, 0, 8 }, // reshape->src[0]  == div
 };

+//node #436 (     UNARY):     ffn_moe_probs-10 ( 256K) [Vulka         ] use=2:    ffn_moe_logits-10 ( 256K) [Vulka         ]
+//node #437 (   RESHAPE): ffn_moe_probs-10 (re ( 256K) [Vulka         ] use=1:     ffn_moe_probs-10 ( 256K) [Vulka         ]
+//node #438 (       ADD): ffn_moe_probs_biased ( 256K) [Vulka         ] use=1:     ffn_moe_probs-10 ( 256K) [Vulka         ] blk.10.exp_probs_b.b (   0K) [Vulka         ]
+//node #439 (   ARGSORT):   ffn_moe_argsort-10 ( 256K) [Vulka         ] use=1: ffn_moe_probs_biased ( 256K) [Vulka         ]
+//node #440 (      VIEW):      ffn_moe_topk-10 ( 255K) [Vulka         ] use=3:   ffn_moe_argsort-10 ( 256K) [Vulka         ]
+//node #441 (  GET_ROWS):   ffn_moe_weights-10 (  12K) [Vulka         ] use=1: ffn_moe_probs-10 (re ( 256K) [Vulka         ]      ffn_moe_topk-10 ( 255K) [Vulka         ]
+//node #442 (   RESHAPE): ffn_moe_weights-10 ( (  12K) [Vulka         ] use=2:   ffn_moe_weights-10 (  12K) [Vulka         ]
+//node #443 (  SUM_ROWS): ffn_moe_weights_sum- (   2K) [Vulka         ] use=1: ffn_moe_weights-10 ( (  12K) [Vulka         ]
+//node #444 (     CLAMP): ffn_moe_weights_sum_ (   2K) [Vulka         ] use=1: ffn_moe_weights_sum- (   2K) [Vulka         ]
+//node #445 (       DIV): ffn_moe_weights_norm (  12K) [Vulka         ] use=1: ffn_moe_weights-10 ( (  12K) [Vulka         ] ffn_moe_weights_sum_ (   2K) [Vulka         ]
+//node #446 (   RESHAPE): ffn_moe_weights_norm (  12K) [Vulka         ] use=1: ffn_moe_weights_norm (  12K) [Vulka         ]
+static constexpr std::initializer_list<std::array<int, 3>> topk_moe_sigmoid_norm_bias_edges {
+    { 1, 0, 0 }, // reshape->src[0]  == sigmoid
+    { 2, 0, 0 }, // add->src[0]      == sigmoid
+    { 3, 0, 2 }, // argsort->src[0]  == add
+    { 4, 0, 3 }, // view->src[0]     == argsort
+    { 5, 0, 1 }, // get_rows->src[0] == reshape
+    { 5, 1, 4 }, // get_rows->src[1] == view
+    { 6, 0, 5 }, // reshape->src[0]  == get_rows
+    { 7, 0, 6 }, // sum_rows->src[0] == reshape
+    { 8, 0, 7 }, // clamp->src[0]    == sum_rows
+    { 9, 0, 6 }, // div->src[0]      == reshape
+    { 9, 1, 8 }, // div->src[1]      == clamp
+    {10, 0, 9 }, // reshape->src[0]  == div
+};
+
 // same as early_softmax_norm but ending after the get_rows
 static constexpr std::initializer_list<std::array<int, 3>> topk_moe_early_softmax_edges {
    { 1, 0, 0 }, // reshape->src[0]  == softmax
@ -491,16 +524,10 @@ enum topk_moe_mode {
    TOPK_MOE_EARLY_SOFTMAX,
    TOPK_MOE_EARLY_SOFTMAX_NORM,
    TOPK_MOE_LATE_SOFTMAX,
+    TOPK_MOE_SIGMOID_NORM_BIAS,
    TOPK_MOE_COUNT,
 };

-static topk_moe_mode ggml_vk_num_additional_ops_to_topk_moe_mode(uint32_t num) {
-    topk_moe_mode mode = num == topk_moe_early_softmax_norm.size() - 1 ? TOPK_MOE_EARLY_SOFTMAX_NORM :
-                         num == topk_moe_early_softmax.size() - 1      ? TOPK_MOE_EARLY_SOFTMAX :
-                                                                         TOPK_MOE_LATE_SOFTMAX;
-    return mode;
-}
-
 static constexpr std::initializer_list<std::array<int, 3>> rope_view_set_rows_edges {
    { 1, 0, 0 }, // view->src[0]     == rope
    { 2, 0, 1 }, // set_rows->src[0] == view
@ -766,7 +793,7 @@ struct vk_device_struct {
    vk_pipeline pipeline_count_experts;

    // [2] is for whether to take n_experts from spec constant (0) or push constant (1)
-    vk_pipeline pipeline_topk_moe[num_topk_moe_pipelines][TOPK_MOE_COUNT][2];
+    vk_pipeline pipeline_topk_moe[num_topk_moe_pipelines][2];

    std::vector<vk_pipeline_ref> all_pipelines;

@ -1181,6 +1208,11 @@ struct vk_op_topk_moe_push_constants {
    uint32_t n_expert_used;
    float clamp_min;
    float clamp_max;
+    uint32_t gating_func;
+    uint32_t has_bias;
+    uint32_t with_norm;
+    float output_scale;
+    float output_bias;
 };

 struct vk_op_add_id_push_constants {
@ -1771,6 +1803,8 @@ struct ggml_backend_vk_context {
    // Bit 'i' means nodes[start_of_fusion + i] writes to memory.
    // If there's no fusion, bit 0 is still set.
    int fused_ops_write_mask {};
+    topk_moe_mode fused_topk_moe_mode {};
+    bool fused_topk_moe_scale {};

    // for GGML_VK_PERF_LOGGER
    std::unique_ptr<vk_perf_logger> perf_logger;
@ -4291,9 +4325,7 @@ static void ggml_vk_load_shaders(vk_device& device) {

    for (uint32_t use_push = 0; use_push < 2; ++use_push) {
        for (uint32_t i = 0; i < num_topk_moe_pipelines; ++i) {
-            ggml_vk_create_pipeline2(device, device->pipeline_topk_moe[i][TOPK_MOE_EARLY_SOFTMAX][use_push],      "topk_moe_f32_early_softmax_"+std::to_string(i),       topk_moe_f32_len, topk_moe_f32_data, "main", 3, sizeof(vk_op_topk_moe_push_constants), {1, 1, 1}, {device->subgroup_size, 1u<<i, 0, 0, use_push}, 1, true, true, device->subgroup_size);
-            ggml_vk_create_pipeline2(device, device->pipeline_topk_moe[i][TOPK_MOE_EARLY_SOFTMAX_NORM][use_push], "topk_moe_f32_early_softmax_norm"+std::to_string(i),   topk_moe_f32_len, topk_moe_f32_data, "main", 3, sizeof(vk_op_topk_moe_push_constants), {1, 1, 1}, {device->subgroup_size, 1u<<i, 1, 0, use_push}, 1, true, true, device->subgroup_size);
-            ggml_vk_create_pipeline2(device, device->pipeline_topk_moe[i][TOPK_MOE_LATE_SOFTMAX][use_push],       "topk_moe_f32_late_softmax"+std::to_string(i),         topk_moe_f32_len, topk_moe_f32_data, "main", 3, sizeof(vk_op_topk_moe_push_constants), {1, 1, 1}, {device->subgroup_size, 1u<<i, 0, 1, use_push}, 1, true, true, device->subgroup_size);
+            ggml_vk_create_pipeline2(device, device->pipeline_topk_moe[i][use_push], "topk_moe_f32_"+std::to_string(i), topk_moe_f32_len, topk_moe_f32_data, "main", 4, sizeof(vk_op_topk_moe_push_constants), {1, 1, 1}, {device->subgroup_size, 1u<<i, use_push}, 1, true, true, device->subgroup_size);
        }
    }

@ -8684,10 +8716,9 @@ static vk_pipeline ggml_vk_op_get_pipeline(ggml_backend_vk_context * ctx, const
        if (ctx->num_additional_fused_ops) {
            uint32_t idx = (uint32_t)ceilf(log2f(float(dst->ne[0])));
            GGML_ASSERT(idx < num_topk_moe_pipelines);
-            topk_moe_mode mode = ggml_vk_num_additional_ops_to_topk_moe_mode(ctx->num_additional_fused_ops);
            // use n_experts from push constant if it's not equal to the power of two spec constant
            bool use_push = dst->ne[0] != (1u << idx);
-            return ctx->device->pipeline_topk_moe[idx][mode][use_push];
+            return ctx->device->pipeline_topk_moe[idx][use_push];
        }

        if (src0->type == GGML_TYPE_F32 && (src1 == nullptr || src1->type == GGML_TYPE_F32) && dst->type == GGML_TYPE_F32) {
@ -10346,14 +10377,16 @@ static void ggml_vk_soft_max_back(ggml_backend_vk_context * ctx, vk_context& sub
 }

 static void ggml_vk_topk_moe(ggml_backend_vk_context * ctx, vk_context& subctx, ggml_cgraph * cgraph, int node_idx) {
-    topk_moe_mode mode = ggml_vk_num_additional_ops_to_topk_moe_mode(ctx->num_additional_fused_ops);
+    topk_moe_mode mode = ctx->fused_topk_moe_mode;
    ggml_tensor * logits = cgraph->nodes[node_idx + 0]->src[0];
-    ggml_tensor * weights = (mode == TOPK_MOE_EARLY_SOFTMAX_NORM) ? cgraph->nodes[node_idx + 9] :
-                            (mode == TOPK_MOE_EARLY_SOFTMAX)      ? cgraph->nodes[node_idx + 4] :
-                                                                    cgraph->nodes[node_idx + 5];
-    ggml_tensor * ids = (mode == TOPK_MOE_LATE_SOFTMAX) ? cgraph->nodes[node_idx + 1] : cgraph->nodes[node_idx + 3];
+    ggml_tensor * bias = (mode == TOPK_MOE_SIGMOID_NORM_BIAS) ? cgraph->nodes[node_idx + 2]->src[1] : logits;
+    ggml_tensor * weights = cgraph->nodes[node_idx + ctx->num_additional_fused_ops];
+    ggml_tensor * ids = (mode == TOPK_MOE_SIGMOID_NORM_BIAS) ? cgraph->nodes[node_idx + 4] :
+                        (mode == TOPK_MOE_LATE_SOFTMAX) ?      cgraph->nodes[node_idx + 1] :
+                                                               cgraph->nodes[node_idx + 3];

    GGML_ASSERT(logits->type == GGML_TYPE_F32);
+    GGML_ASSERT(bias->type == GGML_TYPE_F32);
    GGML_ASSERT(weights->type == GGML_TYPE_F32);
    GGML_ASSERT(ids->type == GGML_TYPE_I32);

@ -10368,6 +10401,7 @@ static void ggml_vk_topk_moe(ggml_backend_vk_context * ctx, vk_context& subctx,
    ggml_pipeline_request_descriptor_sets(ctx, pipeline, 1);

    vk_subbuffer logits_buf = ggml_vk_tensor_subbuffer(ctx, logits);
+    vk_subbuffer bias_buf = ggml_vk_tensor_subbuffer(ctx, bias);
    vk_subbuffer weights_buf = ggml_vk_tensor_subbuffer(ctx, weights);
    vk_subbuffer ids_buf = ggml_vk_tensor_subbuffer(ctx, ids);

@ -10375,18 +10409,45 @@ static void ggml_vk_topk_moe(ggml_backend_vk_context * ctx, vk_context& subctx,
    pc.n_rows = n_rows;
    pc.n_experts_push = n_experts;
    pc.n_expert_used = n_expert_used;
+    pc.clamp_min = -std::numeric_limits<float>::infinity();
+    pc.clamp_max = std::numeric_limits<float>::infinity();
    if (mode == TOPK_MOE_EARLY_SOFTMAX_NORM) {
        ggml_tensor * clamp = cgraph->nodes[node_idx + 7];
+        GGML_ASSERT(clamp->op == GGML_OP_CLAMP);
        pc.clamp_min = ggml_get_op_params_f32(clamp, 0);
        pc.clamp_max = ggml_get_op_params_f32(clamp, 1);
    }
+    if (mode == TOPK_MOE_SIGMOID_NORM_BIAS) {
+        ggml_tensor * clamp = cgraph->nodes[node_idx + 8];
+        GGML_ASSERT(clamp->op == GGML_OP_CLAMP);
+        pc.clamp_min = ggml_get_op_params_f32(clamp, 0);
+        pc.clamp_max = ggml_get_op_params_f32(clamp, 1);
+    }
+
+#define GATING_FUNC_SOFTMAX 0
+#define GATING_FUNC_SIGMOID 1
+#define GATING_FUNC_SOFTMAX_WEIGHT 2
+
+    pc.gating_func = mode == TOPK_MOE_SIGMOID_NORM_BIAS ? GATING_FUNC_SIGMOID :
+                     mode == TOPK_MOE_LATE_SOFTMAX ?      GATING_FUNC_SOFTMAX_WEIGHT :
+                                                          GATING_FUNC_SOFTMAX;
+    pc.has_bias = mode == TOPK_MOE_SIGMOID_NORM_BIAS;
+    pc.with_norm = mode == TOPK_MOE_EARLY_SOFTMAX_NORM || mode == TOPK_MOE_SIGMOID_NORM_BIAS;
+    if (ctx->fused_topk_moe_scale) {
+        GGML_ASSERT(weights->op == GGML_OP_SCALE);
+        pc.output_scale = ggml_get_op_params_f32(weights, 0);
+        pc.output_bias = ggml_get_op_params_f32(weights, 1);
+    } else {
+        pc.output_scale = 1.0f;
+        pc.output_bias = 0.0f;
+    }

    GGML_ASSERT(n_expert_used <= n_experts);

    const uint32_t rows_per_block = 4;
    std::array<uint32_t, 3> elements = { CEIL_DIV(n_rows, rows_per_block), 1, 1 };

-    ggml_vk_dispatch_pipeline(ctx, subctx, pipeline, {logits_buf, weights_buf, ids_buf}, pc, elements);
+    ggml_vk_dispatch_pipeline(ctx, subctx, pipeline, {logits_buf, bias_buf, weights_buf, ids_buf}, pc, elements);
 }

 static void ggml_vk_rope(ggml_backend_vk_context * ctx, vk_context& subctx, const ggml_cgraph * cgraph, int node_idx, bool backprop) {
@ -12128,6 +12189,11 @@ static bool ggml_vk_build_graph(ggml_backend_vk_context * ctx, ggml_cgraph * cgr

        break;
    case GGML_OP_UNARY:
+        if (ctx->fused_topk_moe_mode != TOPK_MOE_COUNT) {
+            ggml_vk_topk_moe(ctx, compute_ctx, cgraph, node_idx);
+            break;
+        }
+
        switch (ggml_get_unary_op(node)) {
        case GGML_UNARY_OP_EXP:
        case GGML_UNARY_OP_SILU:
@ -12175,7 +12241,7 @@ static bool ggml_vk_build_graph(ggml_backend_vk_context * ctx, ggml_cgraph * cgr

        break;
    case GGML_OP_SOFT_MAX:
-        if (ctx->num_additional_fused_ops) {
+        if (ctx->fused_topk_moe_mode != TOPK_MOE_COUNT) {
            ggml_vk_topk_moe(ctx, compute_ctx, cgraph, node_idx);
        } else {
            ggml_vk_soft_max(ctx, compute_ctx, src0, src1, src2, node);
@ -12195,7 +12261,7 @@ static bool ggml_vk_build_graph(ggml_backend_vk_context * ctx, ggml_cgraph * cgr

        break;
    case GGML_OP_ARGSORT:
-        if (ctx->num_additional_fused_ops) {
+        if (ctx->fused_topk_moe_mode != TOPK_MOE_COUNT) {
            ggml_vk_topk_moe(ctx, compute_ctx, cgraph, node_idx);
        } else {
            ggml_vk_argsort(ctx, compute_ctx, src0, node);
@ -13048,6 +13114,24 @@ static bool ggml_vk_can_fuse_topk_moe(ggml_backend_vk_context * ctx, const struc
        get_rows = cgraph->nodes[node_idx + 4];
        argsort = cgraph->nodes[node_idx + 2];
        break;
+    case TOPK_MOE_SIGMOID_NORM_BIAS:
+        softmax = cgraph->nodes[node_idx + 0]; // really sigmoid
+        weights = cgraph->nodes[node_idx + 10];
+        get_rows = cgraph->nodes[node_idx + 5];
+        argsort = cgraph->nodes[node_idx + 3];
+        if (ggml_get_unary_op(softmax) != GGML_UNARY_OP_SIGMOID) {
+            return false;
+        }
+        // bias is expected to be 1D
+        if (ggml_nrows(cgraph->nodes[node_idx + 2]->src[1]) != 1 ||
+            !ggml_is_contiguous(cgraph->nodes[node_idx + 2]->src[1])) {
+            return false;
+        }
+        // sigmoid fusion seems to generate infinities on moltenvk
+        if (ctx->device->driver_id == vk::DriverId::eMoltenvk) {
+            return false;
+        }
+        break;
    case TOPK_MOE_EARLY_SOFTMAX:
        softmax = cgraph->nodes[node_idx + 0];
        weights = cgraph->nodes[node_idx + 4];
@ -13071,26 +13155,28 @@ static bool ggml_vk_can_fuse_topk_moe(ggml_backend_vk_context * ctx, const struc
    probs = probs->src[0];
    ggml_tensor * selection_probs = argsort->src[0];

-    if (probs != selection_probs) {
+    if (probs != selection_probs && mode != TOPK_MOE_SIGMOID_NORM_BIAS) {
        return false;
    }

-    const float * op_params = (const float *)softmax->op_params;
-
-    float scale = op_params[0];
-    float max_bias = op_params[1];
-
    if (!ggml_is_contiguous(softmax->src[0]) || !ggml_is_contiguous(weights)) {
        return false;
    }

-    if (scale != 1.0f || max_bias != 0.0f) {
-        return false;
-    }
+    if (softmax->op == GGML_OP_SOFT_MAX) {
+        const float * op_params = (const float *)softmax->op_params;

-    // don't fuse when masks or sinks are present
-    if (softmax->src[1] || softmax->src[2]) {
-        return false;
+        float scale = op_params[0];
+        float max_bias = op_params[1];
+
+        if (scale != 1.0f || max_bias != 0.0f) {
+            return false;
+        }
+
+        // don't fuse when masks or sinks are present
+        if (softmax->src[1] || softmax->src[2]) {
+            return false;
+        }
    }

    const int n_expert = softmax->ne[0];
@ -13363,6 +13449,8 @@ static ggml_status ggml_backend_vk_graph_compute(ggml_backend_t backend, ggml_cg
            total_mul_mat_bytes += bytes;
        }

+        ctx->fused_topk_moe_mode = TOPK_MOE_COUNT;
+        ctx->fused_topk_moe_scale = false;
        const char *fusion_string {};
        if (!ctx->device->disable_fusion) {
            uint32_t num_adds = ggml_vk_fuse_multi_add(ctx, cgraph, i);
@ -13408,13 +13496,23 @@ static ggml_status ggml_backend_vk_graph_compute(ggml_backend_t backend, ggml_cg
                ctx->num_additional_fused_ops = topk_moe_early_softmax_norm.size() - 1;
                // view of argsort writes to memory
                ctx->fused_ops_write_mask |= 1 << 3;
+                ctx->fused_topk_moe_mode = TOPK_MOE_EARLY_SOFTMAX_NORM;
                fusion_string = "TOPK_MOE_EARLY_SOFTMAX_NORM";
+            } else if (ggml_can_fuse_subgraph(cgraph, i, topk_moe_sigmoid_norm_bias, { i + 4, i + 10 }) &&
+                       ggml_check_edges(cgraph, i, topk_moe_sigmoid_norm_bias_edges) &&
+                       ggml_vk_can_fuse_topk_moe(ctx, cgraph, i, TOPK_MOE_SIGMOID_NORM_BIAS)) {
+                ctx->num_additional_fused_ops = topk_moe_sigmoid_norm_bias.size() - 1;
+                // view of argsort writes to memory
+                ctx->fused_ops_write_mask |= 1 << 4;
+                ctx->fused_topk_moe_mode = TOPK_MOE_SIGMOID_NORM_BIAS;
+                fusion_string = "TOPK_MOE_SIGMOID_NORM_BIAS";
            } else if (ggml_can_fuse_subgraph(cgraph, i, topk_moe_early_softmax, { i + 3, i + 4 }) &&
                       ggml_check_edges(cgraph, i, topk_moe_early_softmax_edges) &&
                       ggml_vk_can_fuse_topk_moe(ctx, cgraph, i, TOPK_MOE_EARLY_SOFTMAX)) {
                ctx->num_additional_fused_ops = topk_moe_early_softmax.size() - 1;
                // view of argsort writes to memory
                ctx->fused_ops_write_mask |= 1 << 3;
+                ctx->fused_topk_moe_mode = TOPK_MOE_EARLY_SOFTMAX;
                fusion_string = "TOPK_MOE_EARLY_SOFTMAX";
            } else if (ggml_can_fuse_subgraph(cgraph, i, topk_moe_late_softmax, { i + 1, i + 5 }) &&
                       ggml_check_edges(cgraph, i, topk_moe_late_softmax_edges) &&
@ -13422,8 +13520,17 @@ static ggml_status ggml_backend_vk_graph_compute(ggml_backend_t backend, ggml_cg
                ctx->num_additional_fused_ops = topk_moe_late_softmax.size() - 1;
                // view of argsort writes to memory
                ctx->fused_ops_write_mask |= 1 << 1;
+                ctx->fused_topk_moe_mode = TOPK_MOE_LATE_SOFTMAX;
                fusion_string = "TOPK_MOE_LATE_SOFTMAX";
            }
+            if (ctx->fused_topk_moe_mode != TOPK_MOE_COUNT) {
+                // Look for an additional scale op to fuse - occurs in deepseek2 and nemotron3 nano.
+                if (ggml_can_fuse_subgraph(cgraph, i + ctx->num_additional_fused_ops - 1, { GGML_OP_DIV, GGML_OP_RESHAPE, GGML_OP_SCALE }, { i + ctx->num_additional_fused_ops + 1 }) ||
+                    ggml_can_fuse_subgraph(cgraph, i + ctx->num_additional_fused_ops, { GGML_OP_GET_ROWS, GGML_OP_SCALE }, { i + ctx->num_additional_fused_ops + 1 })) {
+                    ctx->fused_topk_moe_scale = true;
+                    ctx->num_additional_fused_ops++;
+                }
+            }
        }
        ctx->fused_ops_write_mask |= 1 << ctx->num_additional_fused_ops;

@ -13602,6 +13709,9 @@ static void ggml_vk_graph_optimize(ggml_backend_t backend, struct ggml_cgraph *
        if (keep_pattern(topk_moe_early_softmax_norm)) {
            continue;
        }
+        if (keep_pattern(topk_moe_sigmoid_norm_bias)) {
+            continue;
+        }
        if (keep_pattern(topk_moe_early_softmax)) {
            continue;
        }
@ -13628,6 +13738,7 @@ static void ggml_vk_graph_optimize(ggml_backend_t backend, struct ggml_cgraph *
            }
            // Don't pull forward nodes from fusion patterns
            if (match_pattern(topk_moe_early_softmax_norm, j) ||
+                match_pattern(topk_moe_sigmoid_norm_bias, j) ||
                match_pattern(topk_moe_early_softmax, j) ||
                match_pattern(topk_moe_late_softmax, j)) {
                continue;
--- a/ggml/src/ggml-vulkan/vulkan-shaders/topk_moe.comp
+++ b/ggml/src/ggml-vulkan/vulkan-shaders/topk_moe.comp
@ -7,6 +7,10 @@

 #include "types.glsl"

+#define GATING_FUNC_SOFTMAX 0
+#define GATING_FUNC_SIGMOID 1
+#define GATING_FUNC_SOFTMAX_WEIGHT 2
+
 layout (push_constant) uniform parameter
 {
    uint n_rows;
@ -14,15 +18,18 @@ layout (push_constant) uniform parameter
    uint n_expert_used;
    float clamp_min;
    float clamp_max;
+    uint gating_func;
+    uint has_bias;
+    uint with_norm;
+    float output_scale;
+    float output_bias;
 };

 layout(local_size_x_id = 0, local_size_y = 4, local_size_z = 1) in;

 layout(constant_id = 0) const uint WARP_SIZE = 32;
 layout(constant_id = 1) const uint n_experts_spec = 512;
-layout(constant_id = 2) const bool with_norm = true;
-layout(constant_id = 3) const bool late_softmax = false;
-layout(constant_id = 4) const bool nexperts_use_push = false;
+layout(constant_id = 2) const bool nexperts_use_push = false;

 uint n_experts = nexperts_use_push ? n_experts_push : n_experts_spec;

@ -31,8 +38,9 @@ uint n_experts = nexperts_use_push ? n_experts_push : n_experts_spec;
 const uint experts_per_thread = CEIL_DIV(n_experts_spec, WARP_SIZE);

 layout (binding = 0, std430) readonly buffer Logits {float logits[];};
-layout (binding = 1, std430) writeonly buffer Weights {float weights[];};
-layout (binding = 2, std430) writeonly buffer Ids {uint ids[];};
+layout (binding = 1, std430) readonly buffer BiasProbs {float bias[];};
+layout (binding = 2, std430) writeonly buffer Weights {float weights[];};
+layout (binding = 3, std430) writeonly buffer Ids {uint ids[];};

 const float INFINITY = 1.0 / 0.0;

@ -87,20 +95,40 @@ void main() {
    }

    const uint logits_offset = n_experts * row;
+    const uint bias_offset = 0; // 1D
    const uint weights_offset = n_expert_used * row;
    const uint ids_offset = n_experts * row;
    const uint lane = gl_SubgroupInvocationID;

-    float wt[experts_per_thread];
+    float probs[experts_per_thread];

    [[unroll]]
    for (uint i = 0; i < n_experts; i += WARP_SIZE) {
        const uint expert = i + lane;
-        wt[i / WARP_SIZE] = (n_experts % WARP_SIZE == 0 || expert < n_experts) ? logits[logits_offset + expert] : -INFINITY;
+        probs[i / WARP_SIZE] = (n_experts % WARP_SIZE == 0 || expert < n_experts) ? logits[logits_offset + expert] : -INFINITY;
    }

-    if (!late_softmax) {
-        softmax_warp_inplace(wt, n_experts, lane, nexperts_use_push);
+    if (gating_func == GATING_FUNC_SOFTMAX) {
+        softmax_warp_inplace(probs, n_experts, lane, nexperts_use_push);
+    } else if (gating_func == GATING_FUNC_SIGMOID) {
+        [[unroll]]
+        for (int i = 0; i < experts_per_thread; i++) {
+            probs[i] = 1.f / (1.f + exp(-probs[i]));
+        }
+    }
+
+    float selection_probs[experts_per_thread];
+    if (has_bias != 0) {
+        [[unroll]]
+        for (uint i = 0; i < n_experts; i += WARP_SIZE) {
+            const uint expert = i + lane;
+            selection_probs[i / WARP_SIZE] = (n_experts % WARP_SIZE == 0 || expert < n_experts) ? probs[i / WARP_SIZE] + bias[bias_offset + expert] : -INFINITY;
+        }
+    } else {
+        [[unroll]]
+        for (int i = 0; i < experts_per_thread; i++) {
+            selection_probs[i] = probs[i];
+        }
    }

    // at this point, each thread holds a portion of softmax,
@ -117,14 +145,16 @@ void main() {
    }

    for (int k = 0; k < n_expert_used; k++) {
-        float max_val    = wt[0];
+        float max_val    = probs[0];
+        float max_val_s  = selection_probs[0];
        uint   max_expert = lane;

        [[unroll]]
        for (int i = 1; i < experts_per_thread; i++) {
            const uint expert = lane + i * WARP_SIZE;
-            if ((n_experts % WARP_SIZE == 0 || expert < n_experts) && wt[i] > max_val) {
-                max_val    = wt[i];
+            if ((n_experts % WARP_SIZE == 0 || expert < n_experts) && selection_probs[i] > max_val_s) {
+                max_val    = probs[i];
+                max_val_s  = selection_probs[i];
                max_expert = expert;
            }
        }
@ -132,9 +162,11 @@ void main() {
        [[unroll]]
        for (uint mask = WARP_SIZE / 2; mask > 0; mask /= 2) {
            const float val    = subgroupShuffleXor(max_val, mask);
+            const float val_s  = subgroupShuffleXor(max_val_s, mask);
            const uint  expert = subgroupShuffleXor(max_expert, mask);
-            if (val > max_val || (val == max_val && expert < max_expert)) {
+            if (val_s > max_val_s || (val_s == max_val_s && expert < max_expert)) {
                max_val    = val;
+                max_val_s  = val_s;
                max_expert = expert;
            }
        }
@ -144,16 +176,14 @@ void main() {
        }

        if ((max_expert & (WARP_SIZE - 1)) == lane) {
-            wt[max_expert / WARP_SIZE] = -INFINITY;
+            selection_probs[max_expert / WARP_SIZE] = -INFINITY;

            ids[ids_offset + k] = max_expert;
-            if (with_norm) {
-                wt_sum += max_val;
-            }
+            wt_sum += max_val;
        }
    }

-    if (with_norm) {
+    if (with_norm != 0) {
        wt_sum              = subgroupAdd(wt_sum);
        wt_sum              = clamp(wt_sum, clamp_min, clamp_max);
        const float inv_sum = 1.0f / wt_sum;
@ -164,7 +194,7 @@ void main() {
        }
    }

-    if (late_softmax) {
+    if (gating_func == GATING_FUNC_SOFTMAX_WEIGHT) {
        softmax_warp_inplace(output_weights, n_expert_used, lane, true);
    }

@ -172,7 +202,7 @@ void main() {
    for (uint i = 0; i < experts_per_thread; ++i) {
        uint idx = i * WARP_SIZE + lane;
        if (idx < n_expert_used) {
-            weights[weights_offset + idx] = output_weights[i];
+            weights[weights_offset + idx] = output_scale * output_weights[i] + output_bias;
        }
    }
 }
--- a/gguf-py/gguf/constants.py
+++ b/gguf-py/gguf/constants.py
@ -3492,6 +3492,7 @@ class VisionProjectorType:
    COGVLM = "cogvlm"
    JANUS_PRO = "janus_pro"
    LFM2A = "lfm2a" # audio
+    MUSIC_FLAMINGO = "musicflamingo" # audio
    GLM4V = "glm4v"


--- a/scripts/sync-ggml.last
+++ b/scripts/sync-ggml.last
@ -1 +1 @@
-130bc125a88bb57664b88932c48c38a1cb316fac
+ebc3a0f4a56be1c9424a89fbec09962ac34fde85
--- a/src/llama-mmap.cpp
+++ b/src/llama-mmap.cpp
@ -240,9 +240,10 @@ struct llama_file::impl {
                throw std::runtime_error("unexpectedly reached end of file");
            }
        } else {
-            bool successful = false;
-            while (!successful) {
-                off_t ret = read(fd, ptr, len);
+            size_t bytes_read = 0;
+            while (bytes_read < len) {
+                const size_t to_read = len - bytes_read;
+                ssize_t ret = ::read(fd, reinterpret_cast<char *>(ptr) + bytes_read, to_read);

                if (ret == -1) {
                    if (errno == EINTR) {
@ -251,10 +252,16 @@ struct llama_file::impl {
                    throw std::runtime_error(format("read error: %s", strerror(errno)));
                }
                if (ret == 0) {
+                    // EOF: allow if this read was only pulling alignment padding past file end
+                    off_t pos = lseek(fd, 0, SEEK_CUR);
+                    if (pos != -1 && (size_t) pos == size) {
+                        std::memset(reinterpret_cast<char *>(ptr) + bytes_read, 0, len - bytes_read);
+                        return;
+                    }
                    throw std::runtime_error("unexpectedly reached end of file");
                }

-                successful = true;
+                bytes_read += (size_t) ret;
            }
        }
    }
--- a/tests/test-backend-ops.cpp
+++ b/tests/test-backend-ops.cpp
@ -1158,6 +1158,7 @@ struct test_case {
    }

    virtual bool run_whole_graph() { return false; }
+    virtual std::vector<ggml_tensor *> fusion_test_nodes() { return {}; }

    ggml_cgraph * gf = nullptr;
    ggml_cgraph * gb = nullptr;
@ -1391,7 +1392,13 @@ struct test_case {
            GGML_UNUSED(index);
        };

-        const bool cmp_ok = ggml_backend_compare_graph_backend(backend1, backend2, gf, callback, &ud, run_whole_graph() ? out : nullptr);
+        std::vector<ggml_tensor *> fused_nodes_to_verify = fusion_test_nodes();
+        if (fused_nodes_to_verify.size() == 0 && run_whole_graph()) {
+            fused_nodes_to_verify.push_back(out);
+        }
+        const bool cmp_ok = ggml_backend_compare_graph_backend(backend1, backend2, gf, callback, &ud,
+                                                               run_whole_graph() ? fused_nodes_to_verify.data() : nullptr,
+                                                               fused_nodes_to_verify.size());

        ggml_backend_buffer_free(buf);

@ -5180,6 +5187,8 @@ struct test_topk_moe : public test_case {
    const bool bias_probs;
    const MoeGatingFunc gating_func;
    const float scale_w;
+    ggml_tensor * weights {};
+    ggml_tensor * selected_experts {};

    test_topk_moe(std::array<int64_t, 4> ne              = { 10, 5, 1, 1 },
                  int                    n_expert_used   = 1,
@ -5217,16 +5226,16 @@ struct test_topk_moe : public test_case {

        ggml_tensor * selection_probs = probs;
        if (bias_probs) {
-            ggml_tensor * exp_probs_b = ggml_new_tensor(ctx, GGML_TYPE_F32, 4, ne.data());
+            ggml_tensor * exp_probs_b = ggml_new_tensor_1d(ctx, GGML_TYPE_F32, ne[0]);
            ggml_set_name(exp_probs_b, "exp_probs_b");
            selection_probs = ggml_add(ctx, probs, exp_probs_b);
            ggml_set_name(selection_probs, "selection_probs");
        }

-        ggml_tensor * selected_experts = ggml_argsort_top_k(ctx, selection_probs, n_expert_used); // [n_expert_used, n_tokens]
+        selected_experts = ggml_argsort_top_k(ctx, selection_probs, n_expert_used); // [n_expert_used, n_tokens]
        ggml_set_name(selected_experts, "selected_experts");

-        ggml_tensor * weights = ggml_get_rows(ctx, ggml_reshape_3d(ctx, probs, 1, n_expert, n_tokens), selected_experts); // [1, n_expert_used, n_tokens]
+        weights = ggml_get_rows(ctx, ggml_reshape_3d(ctx, probs, 1, n_expert, n_tokens), selected_experts); // [1, n_expert_used, n_tokens]
        ggml_set_name(weights, "weights");

        if (gating_func == GATING_FUNC_SOFTMAX_WEIGHT) {
@ -5252,6 +5261,21 @@ struct test_topk_moe : public test_case {
        ggml_set_name(weights, "weights");
        return weights;
    }
+    // Verify two outputs
+    std::vector<ggml_tensor *> fusion_test_nodes() override { return { selected_experts, weights }; }
+
+    // allow output in arbitrary order
+    double err(const float * a, const float * b, size_t n) override {
+        std::vector<float> a2(n);
+        std::vector<float> b2(n);
+        for (size_t i = 0; i < n; ++i) {
+            a2[i] = a[i];
+            b2[i] = b[i];
+        }
+        std::sort(a2.begin(), a2.end());
+        std::sort(b2.begin(), b2.end());
+        return nmse(a2.data(), b2.data(), n);
+    }
 };

 struct test_mul_mat_vec_fusion : public test_case {
--- a/tests/test-chat.cpp
+++ b/tests/test-chat.cpp
@ -724,6 +724,30 @@ static void test_tools_oaicompat_json_conversion() {
            "]"
        ),
        common_chat_tools_to_json_oaicompat<json>({special_function_tool}).dump(2));
+
+    {
+        auto tools_no_params = common_chat_tools_parse_oaicompat(json::parse(
+            R"([{"type": "function", "function": {"name": "test_func", "description": "A test"}}])"));
+        assert_equals((size_t) 1, tools_no_params.size());
+        assert_equals(std::string("test_func"), tools_no_params[0].name);
+        assert_equals(std::string("A test"), tools_no_params[0].description);
+        assert_equals(std::string("{}"), tools_no_params[0].parameters);
+    }
+    {
+        auto tools_no_desc = common_chat_tools_parse_oaicompat(json::parse(
+            R"([{"type": "function", "function": {"name": "test_func", "parameters": {"type": "object"}}}])"));
+        assert_equals((size_t) 1, tools_no_desc.size());
+        assert_equals(std::string("test_func"), tools_no_desc[0].name);
+        assert_equals(std::string(""), tools_no_desc[0].description);
+    }
+    {
+        auto tools_minimal = common_chat_tools_parse_oaicompat(json::parse(
+            R"([{"type": "function", "function": {"name": "test_func"}}])"));
+        assert_equals((size_t) 1, tools_minimal.size());
+        assert_equals(std::string("test_func"), tools_minimal[0].name);
+        assert_equals(std::string(""), tools_minimal[0].description);
+        assert_equals(std::string("{}"), tools_minimal[0].parameters);
+    }
 }

 static void test_template_output_parsers() {
--- a/tools/mtmd/clip-impl.h
+++ b/tools/mtmd/clip-impl.h
@ -180,6 +180,7 @@ enum projector_type {
    PROJECTOR_TYPE_GLMA,
    PROJECTOR_TYPE_QWEN25O, // will be replaced by QWEN2A or QWEN25VL depending on clip_ctx
    PROJECTOR_TYPE_VOXTRAL,
+    PROJECTOR_TYPE_MUSIC_FLAMINGO,
    PROJECTOR_TYPE_LFM2,
    PROJECTOR_TYPE_KIMIVL,
    PROJECTOR_TYPE_LIGHTONOCR,
@ -209,6 +210,7 @@ static std::map<projector_type, std::string> PROJECTOR_TYPE_NAMES = {
    { PROJECTOR_TYPE_GLMA,      "glma"},
    { PROJECTOR_TYPE_QWEN25O,   "qwen2.5o"},
    { PROJECTOR_TYPE_VOXTRAL,   "voxtral"},
+    { PROJECTOR_TYPE_MUSIC_FLAMINGO, "musicflamingo"},
    { PROJECTOR_TYPE_LFM2,      "lfm2"},
    { PROJECTOR_TYPE_KIMIVL,    "kimivl"},
    { PROJECTOR_TYPE_LIGHTONOCR,"lightonocr"},
--- a/tools/mtmd/clip-model.h
+++ b/tools/mtmd/clip-model.h
@ -319,7 +319,8 @@ struct clip_model {

    bool audio_has_avgpool() const {
        return proj_type == PROJECTOR_TYPE_QWEN2A
-            || proj_type == PROJECTOR_TYPE_VOXTRAL;
+            || proj_type == PROJECTOR_TYPE_VOXTRAL
+            || proj_type == PROJECTOR_TYPE_MUSIC_FLAMINGO;
    }

    bool audio_has_stack_frames() const {
--- a/tools/mtmd/clip.cpp
+++ b/tools/mtmd/clip.cpp
@ -818,6 +818,7 @@ static ggml_cgraph * clip_image_build_graph(clip_ctx * ctx, const clip_image_f32
        case PROJECTOR_TYPE_VOXTRAL:
        case PROJECTOR_TYPE_QWEN2A:
        case PROJECTOR_TYPE_GLMA:
+        case PROJECTOR_TYPE_MUSIC_FLAMINGO:
            {
                builder = std::make_unique<clip_graph_whisper_enc>(ctx, img);
            } break;
@ -1176,6 +1177,7 @@ struct clip_model_loader {
                case PROJECTOR_TYPE_QWEN2A:
                case PROJECTOR_TYPE_GLMA:
                case PROJECTOR_TYPE_VOXTRAL:
+                case PROJECTOR_TYPE_MUSIC_FLAMINGO:
                    {
                        bool require_stack = model.proj_type == PROJECTOR_TYPE_ULTRAVOX ||
                                             model.proj_type == PROJECTOR_TYPE_VOXTRAL ||
@ -1576,6 +1578,17 @@ struct clip_model_loader {
                    model.mm_1_w = get_tensor(string_format(TN_MM_AUDIO_MLP, 1, "weight"));
                    model.mm_2_w = get_tensor(string_format(TN_MM_AUDIO_MLP, 2, "weight"));
                } break;
+            case PROJECTOR_TYPE_MUSIC_FLAMINGO:
+                {
+                    model.conv1d_1_w = get_tensor(string_format(TN_CONV1D, 1, "weight"));
+                    model.conv1d_1_b = get_tensor(string_format(TN_CONV1D, 1, "bias"));
+                    model.conv1d_2_w = get_tensor(string_format(TN_CONV1D, 2, "weight"));
+                    model.conv1d_2_b = get_tensor(string_format(TN_CONV1D, 2, "bias"));
+                    model.mm_1_w = get_tensor(string_format(TN_MM_AUDIO_MLP, 1, "weight"));
+                    model.mm_1_b = get_tensor(string_format(TN_MM_AUDIO_MLP, 1, "bias"));
+                    model.mm_2_w = get_tensor(string_format(TN_MM_AUDIO_MLP, 2, "weight"));
+                    model.mm_2_b = get_tensor(string_format(TN_MM_AUDIO_MLP, 2, "bias"));
+                } break;
            case PROJECTOR_TYPE_INTERNVL:
                {
                    model.mm_0_w = get_tensor(string_format(TN_MVLM_PROJ_MLP, 0, "weight"));
@ -3031,6 +3044,7 @@ int clip_n_output_tokens(const struct clip_ctx * ctx, struct clip_image_f32 * im
        case PROJECTOR_TYPE_VOXTRAL:
        case PROJECTOR_TYPE_ULTRAVOX:
        case PROJECTOR_TYPE_QWEN2A:
+        case PROJECTOR_TYPE_MUSIC_FLAMINGO:
            {
                n_patches = img->nx;

@ -3403,6 +3417,7 @@ bool clip_image_batch_encode(clip_ctx * ctx, const int n_threads, const clip_ima
        case PROJECTOR_TYPE_ULTRAVOX:
        case PROJECTOR_TYPE_LFM2:
        case PROJECTOR_TYPE_VOXTRAL:
+        case PROJECTOR_TYPE_MUSIC_FLAMINGO:
        case PROJECTOR_TYPE_JANUS_PRO:
        case PROJECTOR_TYPE_COGVLM:
            {
@ -3526,6 +3541,7 @@ int clip_n_mmproj_embd(const struct clip_ctx * ctx) {
            return ctx->model.projection->ne[1];
        case PROJECTOR_TYPE_ULTRAVOX:
        case PROJECTOR_TYPE_VOXTRAL:
+        case PROJECTOR_TYPE_MUSIC_FLAMINGO:
            return ctx->model.mm_2_w->ne[1];
        case PROJECTOR_TYPE_INTERNVL:
            return ctx->model.mm_3_w->ne[1];
@ -3587,7 +3603,8 @@ bool clip_has_whisper_encoder(const struct clip_ctx * ctx) {
    return ctx->proj_type() == PROJECTOR_TYPE_ULTRAVOX
        || ctx->proj_type() == PROJECTOR_TYPE_QWEN2A
        || ctx->proj_type() == PROJECTOR_TYPE_GLMA
-        || ctx->proj_type() == PROJECTOR_TYPE_VOXTRAL;
+        || ctx->proj_type() == PROJECTOR_TYPE_VOXTRAL
+        || ctx->proj_type() == PROJECTOR_TYPE_MUSIC_FLAMINGO;
 }

 bool clip_encode_float_image (struct clip_ctx * ctx, int n_threads, float * img, int h, int w, float * vec) {
--- a/tools/mtmd/models/whisper-enc.cpp
+++ b/tools/mtmd/models/whisper-enc.cpp
@ -86,6 +86,15 @@ ggml_cgraph * clip_graph_whisper_enc::build() {
            FFN_GELU_ERF,
            -1);

+    } else if (proj_type == PROJECTOR_TYPE_MUSIC_FLAMINGO) {
+        // projector
+        cur = build_ffn(cur,
+            model.mm_1_w, model.mm_1_b,
+            nullptr, nullptr,
+            model.mm_2_w, model.mm_2_b,
+            FFN_GELU_ERF,
+            -1);
+
    } else if (proj_type == PROJECTOR_TYPE_GLMA) {
            cur = ggml_norm(ctx0, cur, hparams.eps);
            cur = ggml_mul(ctx0, cur, model.mm_norm_pre_w);
--- a/tools/mtmd/mtmd.cpp
+++ b/tools/mtmd/mtmd.cpp
@ -330,6 +330,7 @@ struct mtmd_context {
            case PROJECTOR_TYPE_ULTRAVOX:
            case PROJECTOR_TYPE_VOXTRAL:
            case PROJECTOR_TYPE_GLMA:
+            case PROJECTOR_TYPE_MUSIC_FLAMINGO:
                audio_preproc = std::make_unique<mtmd_audio_preprocessor_whisper>(ctx_a);
                break;
            case PROJECTOR_TYPE_LFM2A:
@ -352,6 +353,9 @@ struct mtmd_context {
            // [BEGIN_AUDIO] ... (embeddings) ...
            aud_beg = "[BEGIN_AUDIO]";

+        } else if (proj == PROJECTOR_TYPE_MUSIC_FLAMINGO) {
+            // <sound> ... (embeddings) ...
+            aud_beg = "<sound>";
        }
    }

--- a/tools/quantize/quantize.cpp
+++ b/tools/quantize/quantize.cpp
@ -12,6 +12,7 @@
 #include <cmath>
 #include <cctype>
 #include <algorithm>
+#include <filesystem>

 struct quant_option {
    std::string name;
@ -643,6 +644,11 @@ int main(int argc, char ** argv) {
        return 1;
    }

+    if (std::error_code ec; std::filesystem::equivalent(fname_inp, fname_out, ec)) {
+        fprintf(stderr, "%s: error: input and output files are the same: '%s'\n", __func__, fname_inp.c_str());
+        return 1;
+    }
+
    print_build_info();

    fprintf(stderr, "%s: quantizing '%s' to '%s' as %s", __func__, fname_inp.c_str(), fname_out.c_str(), ftype_str.c_str());
--- a/tools/server/public/index.html.gz
+++ b/tools/server/public/index.html.gz
--- a/tools/server/webui/src/lib/utils/clipboard.ts
+++ b/tools/server/webui/src/lib/utils/clipboard.ts
@ -65,10 +65,7 @@ export async function copyCodeToClipboard(
 	successMessage = 'Code copied to clipboard',
 	errorMessage = 'Failed to copy code'
 ): Promise<boolean> {
-	const doc = new DOMParser().parseFromString(rawCode, 'text/html');
-	const decodedCode = doc.body.textContent ?? rawCode;
-
-	return copyToClipboard(decodedCode, successMessage, errorMessage);
+	return copyToClipboard(rawCode, successMessage, errorMessage);
 }

 /**
Author	SHA1	Message	Date
Anri Lombard	d5574c919c	webui: fix code copy stripping XML/HTML tags (#18518 ) * webui: fix code copy stripping XML/HTML tags * webui: update static build	2026-01-01 13:44:11 +01:00
Aman Gupta	26831bded9	ggml-cuda: remove unneccesary prints on ggml_cuda_init (#18502 )	2026-01-01 19:18:43 +08:00
Jeff Bolz	be47fb9285	vulkan: extend topk_moe to handle sigmoid w/exp_probs_b for nemotron (#18295 ) * vulkan: extend topk_moe to handle sigmoid w/exp_probs_b for nemotron Also handle GGML_OP_SCALE at the end (nemotron, deepseek2). Fewer pipeline variants and spec constants, just use push constants. In test_topk_moe, change exp_probs_b to be 1D, matching real networks. Update test-backend-ops and ggml-backend to allow verifying multiple outputs in a fusion test (topk_moe has two outputs). Previously only the final node was verified. * change test_topk_moe to allow results in arbitrary order * disable sigmoid fusion for moltenvk	2026-01-01 08:58:27 +01:00
triplenom	9e10bd2eaf	llama: handle short reads in direct I/O path (#18504 )	2026-01-01 10:24:43 +08:00
Anri Lombard	4cd162a123	chat: make tool description and parameters optional per OpenAI spec (#18478 ) * chat: make tool description and parameters optional per OpenAI spec Per the OpenAI API specification, both 'description' and 'parameters' fields in tool function definitions are optional. Previously, the parser would throw an exception if these fields were missing. Attempts to fix #17667 * refactor: use value() for cleaner optional field access	2025-12-31 17:21:37 -06:00
Georgi Gerganov	13814eb370	sync : ggml	2025-12-31 18:54:43 +02:00
Georgi Gerganov	54f67b9b66	ggml : bump version to 0.9.5 (ggml/1410)	2025-12-31 18:54:43 +02:00
Anri Lombard	33ded988ba	quantize: prevent input/output file collision (#18451 ) Check if input and output files are the same before quantizing to prevent file corruption when mmap reads from a file being written to. Fixes #12753	2025-12-31 23:29:03 +08:00
Sigbjørn Skjæret	0db8109849	convert : lint fix (#18507 )	2025-12-31 14:28:21 +01:00
Henry147147	9b8329de7a	mtmd : Adding support for Nvidia Music Flamingo Model (#18470 ) * Inital commit, debugging q5_k_s quant * Made hf_to_gguf extend whisper to reduce code duplication * addressed convert_hf_to_gguf pull request issue --------- Co-authored-by: Henry D <henrydorsey147@gmail.com>	2025-12-31 12:13:23 +01:00
gatbontonpc	9a6369bb60	metal : add count_equal op (#18314 ) * add count equal for metal * remove trailing whitespace * updated doc ops table * changed shmem to i32 * added multi tg and templating * removed BLAS support from Metal docs * Apply suggestions from code review Co-authored-by: Georgi Gerganov <ggerganov@gmail.com> * add memset to set dst to 0 * metal : cleanup --------- Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>	2025-12-31 10:39:48 +02:00
Johannes Gäßler	ecc343de63	CUDA: fix KQ max calculation (#18487 )	2025-12-31 09:37:00 +01:00
Georgi Gerganov	01ade96e71	metal : remove BF16 x F16 kernels (#18456 )	2025-12-31 09:53:48 +02:00