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

Revert changes to KimiVLForConditionalGeneration

This commit is contained in:
Aes Sedai 2026-02-01 02:19:53 -08:00
parent 9c44981c01
commit 9b14cb8b28
1 changed files with 1 additions and 9 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

10
convert_hf_to_gguf.py
View File

@ -10705,17 +10705,9 @@ class KimiVLModel(MmprojModel):
self.gguf_writer.add_vision_attention_layernorm_eps(self.hparams_vision.get("layer_norm_eps", 1e-5))
def modify_tensors(self, data_torch: Tensor, name: str, bid: int | None) -> Iterable[tuple[str, Tensor]]:
is_vision_tensor = "vision_tower" in name or "multi_modal_projector" in name or "mm_projector" in name
is_vision_tensor = "vision_tower" in name or "multi_modal_projector" in name
if is_vision_tensor:
# update names:
# "mm_projector.proj.0" -> "mm_projector.proj.linear_1.",
# "mm_projector.proj.2" -> "mm_projector.proj.linear_2.",
if "proj.0." in name:
name = name.replace(".0.", ".linear_1.")
if "proj.2." in name:
name = name.replace(".2.", ".linear_2.")
if "pos_emb.weight" in name:
data_torch = data_torch.view(data_torch.shape[0] * data_torch.shape[1], data_torch.shape[2])