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

convert: quick fix

This commit is contained in:
bluebread 2025-12-17 03:58:31 +00:00
parent 616f009e28
commit e5d426be98
1 changed files with 2 additions and 4 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

6
convert_hf_to_gguf.py
View File

@ -7256,13 +7256,11 @@ class DeepseekV2Model(TextModel):
if is_ocr: if is_ocr:
self.hparams['rope_theta'] = self.hparams.get('rope_theta', 10000.0) self.hparams['rope_theta'] = self.hparams.get('rope_theta', 10000.0)
self.hparams['rms_norm_eps'] = self.hparams.get('rms_norm_eps', 1e-6)
else: else:
# note: deepseek2 using MLA converts into MQA (ie: GQA with 1 group) # note: deepseek2 using MLA converts into MQA (ie: GQA with 1 group)
self.hparams["num_key_value_heads"] = 1 self.hparams["num_key_value_heads"] = 1
if (rope_mscale_all := self.rope_parameters.get("mscale_all_dim")) is not None: self.hparams['rms_norm_eps'] = self.hparams.get('rms_norm_eps', 1e-6)
self.hparams["rms_norm_eps"] = self.hparams.get("rms_norm_eps", 1e-6)
super().set_gguf_parameters() super().set_gguf_parameters()
hparams = self.hparams hparams = self.hparams