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

convert : support Glm4MoeLite (#18936) 

* initial commit for branch

* add glm-4.7-flash, move tokenizer hash

* use `glm4` pretok

* silence flake8 E302 (CI)

* apply review feedback

* add <|user|> as eog

* also add EOG `<|observation|>`

* revert llama-vocab

* inherit vocab from glm4

---------

Co-authored-by: Xuan Son Nguyen <son@huggingface.co>
This commit is contained in:
ddh0 2026-01-19 16:09:20 -06:00 committed by GitHub
parent 959ecf7f23
commit 1706a6d7c6
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
2 changed files with 31 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

31
convert_hf_to_gguf.py
View File

@ -1078,6 +1078,9 @@ class TextModel(ModelBase):
if chkhsh == "b3d1dd861f1d4c5c0d2569ce36baf3f90fe8a102db3de50dd71ff860d91be3df":
# ref: https://huggingface.co/aari1995/German_Semantic_V3
res = "jina-v2-de"
if chkhsh == "cdf5f35325780597efd76153d4d1c16778f766173908894c04afc20108536267":
# ref: https://huggingface.co/zai-org/GLM-4.7-Flash
res = "glm4"
if chkhsh == "0ef9807a4087ebef797fc749390439009c3b9eda9ad1a097abbe738f486c01e5":
# ref: https://huggingface.co/meta-llama/Meta-Llama-3-8B
res = "llama-bpe"
@ -7458,7 +7461,7 @@ class DeepseekModel(TextModel):
"DeepseekV3ForCausalLM",
"KimiVLForConditionalGeneration",
"YoutuForCausalLM",
"YoutuVLForConditionalGeneration"
"YoutuVLForConditionalGeneration",
)
class DeepseekV2Model(TextModel):
model_arch = gguf.MODEL_ARCH.DEEPSEEK2
@ -8446,6 +8449,32 @@ class Glm4MoeModel(TextModel):
raise ValueError(f"Unprocessed experts: {experts}")
@ModelBase.register("Glm4MoeLiteForCausalLM")
class Glm4MoeLiteModel(DeepseekV2Model):
model_arch = gguf.MODEL_ARCH.DEEPSEEK2
# copied from Glm4MoeModel
def set_vocab(self):
from transformers import AutoTokenizer
tokenizer = AutoTokenizer.from_pretrained(self.dir_model)
special_vocab = gguf.SpecialVocab(self.dir_model, load_merges=True)
tokens, toktypes, tokpre = self.get_vocab_base()
self.gguf_writer.add_tokenizer_model("gpt2")
self.gguf_writer.add_tokenizer_pre(tokpre)
self.gguf_writer.add_token_list(tokens)
self.gguf_writer.add_token_types(toktypes)
# Special tokens
# Note: Using <|endoftext|> (151329) for eot causes endless generation
special_vocab._set_special_token("bos", tokenizer.get_added_vocab()["[gMASK]"]) # 151331
special_vocab._set_special_token("eot", tokenizer.get_added_vocab()["<|user|>"]) # 151336
special_vocab._set_special_token("unk", tokenizer.get_added_vocab()["<|endoftext|>"]) # 151329
special_vocab._set_special_token("eom", tokenizer.get_added_vocab()["<|observation|>"]) # 151338
special_vocab.add_to_gguf(self.gguf_writer)
@ModelBase.register("GlmForCausalLM", "ChatGLMModel", "ChatGLMForConditionalGeneration")
class ChatGLMModel(TextModel):
model_arch = gguf.MODEL_ARCH.CHATGLM

1
convert_hf_to_gguf_update.py
View File

@ -170,6 +170,7 @@ pre_computed_hashes = [
{"name": "grok-2", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/alvarobartt/grok-2-tokenizer", "chkhsh": "66b8d4e19ab16c3bfd89bce5d785fb7e0155e8648708a1f42077cb9fe002c273"},
# jina-v2-de variants
{"name": "jina-v2-de", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/aari1995/German_Semantic_V3", "chkhsh": "b3d1dd861f1d4c5c0d2569ce36baf3f90fe8a102db3de50dd71ff860d91be3df"},
{"name": "glm4", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/zai-org/GLM-4.7-Flash", "chkhsh": "cdf5f35325780597efd76153d4d1c16778f766173908894c04afc20108536267"},
]