happyz
/
llama.cpp
mirror of https://github.com/ggerganov/llama.cpp.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

refactor: clean up VaetkiModel class

This commit is contained in:
suhyun-hwang 2026-01-14 23:50:10 +09:00
parent ad04d34047
commit 5d0870207a
1 changed files with 0 additions and 12 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

12
convert_hf_to_gguf.py
View File

@ -7711,14 +7711,6 @@ class VaetkiModel(TextModel):
self.gguf_writer.add_add_space_prefix(False)
def __init__(self, *args, **kwargs):
super().__init__(*args, **kwargs)
# Set rope_parameters for hybrid attention (transformers 5.0 format)
self.rope_parameters = {
"full_attention": {"rope_theta": self.hparams.get("rope_theta_global", 1000000.0)},
"sliding_attention": {"rope_theta": self.hparams.get("rope_theta", 10000.0)}
}
def set_gguf_parameters(self):
super().set_gguf_parameters()
@ -7777,10 +7769,6 @@ class VaetkiModel(TextModel):
if name.startswith("model.vision_model.") or name.startswith("vision_model."):
return []
# Handle lm_head.weight (VAETKI does not use tied embeddings)
if name == "lm_head.weight":
return [(self.format_tensor_name(gguf.MODEL_TENSOR.OUTPUT), data_torch)]
# Remove language_model prefix
if name.startswith("model.language_model."):
name = name.replace("model.language_model.", "model.")