Skip to content

support Glm4MoeLite #18936

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
ngxson merged 9 commits into from
Jan 19, 2026
Merged

support Glm4MoeLite #18936

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
9 commits
Select commit Hold shift + click to select a range
d32969d
initial commit for branch
ddh0 Jan 19, 2026
38e4882
add glm-4.7-flash, move tokenizer hash
ddh0 Jan 19, 2026
eb630d4
use `glm4` pretok
ddh0 Jan 19, 2026
c64f9e0
silence flake8 E302 (CI)
ddh0 Jan 19, 2026
354e2b5
apply review feedback
ddh0 Jan 19, 2026
27961db
add <|user|> as eog
ngxson Jan 19, 2026
4d9befc
also add EOG `<|observation|>`
ddh0 Jan 19, 2026
9ce75e4
revert llama-vocab
ngxson Jan 19, 2026
a15dbef
inherit vocab from glm4
ngxson Jan 19, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
31 changes: 30 additions & 1 deletion convert_hf_to_gguf.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1078,6 +1078,9 @@ def get_vocab_base_pre(self, tokenizer) -> str:
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"
Expand Down Expand Up @@ -7458,7 +7461,7 @@ def prepare_tensors(self):
"DeepseekV3ForCausalLM",
"KimiVLForConditionalGeneration",
"YoutuForCausalLM",
"YoutuVLForConditionalGeneration"
"YoutuVLForConditionalGeneration",
)
class DeepseekV2Model(TextModel):
model_arch = gguf.MODEL_ARCH.DEEPSEEK2
Expand Down Expand Up @@ -8446,6 +8449,32 @@ def prepare_tensors(self):
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
Expand Down
1 change: 1 addition & 0 deletions convert_hf_to_gguf_update.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -170,6 +170,7 @@ class TOKENIZER_TYPE(IntEnum):
{"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"},
]


Expand Down