tqcq/llama.cpp
0
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2025-06-27 20:05:20 +00:00
Code Issues Packages Projects Releases Wiki Activity

convert : BailingMoE : fix qkv split when head_dim is 0 (#12687)

Browse Source 
NOTE: Ling-lite-base is broken, see https://huggingface.co/inclusionAI/Ling-lite-base/discussions/2
This commit is contained in:
Sigbjørn Skjæret
2025-04-01 14:37:13 +02:00
committed by GitHub
parent 3fd072a540
commit 5936a616e4
1 changed files with 2 additions and 5 deletions
Download Patch File Download Diff File Expand all files Collapse all files

7
convert_hf_to_gguf.py
View File

@ -5146,10 +5146,7 @@ class BailingMoeModel(Model):
def set_gguf_parameters(self):
super().set_gguf_parameters()
hparams = self.hparams
if hparams.get("head_dim"):
rope_dim = hparams["head_dim"]
else:
rope_dim = hparams["hidden_size"] // hparams["num_attention_heads"]
rope_dim = hparams.get("head_dim") or hparams["hidden_size"] // hparams["num_attention_heads"]
self.gguf_writer.add_rope_dimension_count(rope_dim)
self.gguf_writer.add_rope_scaling_type(gguf.RopeScalingType.NONE)
@ -5175,7 +5172,7 @@ class BailingMoeModel(Model):
n_head = self.hparams["num_attention_heads"]
n_kv_head = self.hparams.get("num_key_value_heads")
n_embd = self.hparams["hidden_size"]
head_dim = self.hparams.get("head_dim", n_embd // n_head)
head_dim = self.hparams.get("head_dim") or n_embd // n_head
output_name = self.format_tensor_name(gguf.MODEL_TENSOR.OUTPUT)