replica
/
llama.cpp
mirror of https://github.com/ggml-org/llama.cpp.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

convert : fix Ernie4.5 MoE without shared experts (#14746)

This commit is contained in:
Piotr Wilkin (ilintar) 2025-07-18 01:17:16 +02:00 committed by GitHub
parent 760b4484e3
commit 670e1360cd
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
1 changed files with 4 additions and 3 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

7
convert_hf_to_gguf.py
View File

@ -2910,11 +2910,12 @@ class Ernie4_5MoeModel(Ernie4_5Model):
self.gguf_writer.add_expert_used_count(self.hparams["moe_k"])
self.gguf_writer.add_interleave_moe_layer_step(self.hparams["moe_layer_interval"])
self.gguf_writer.add_leading_dense_block_count(self.hparams["moe_layer_start_index"])
self.gguf_writer.add_rope_freq_base(self.hparams["rope_theta"])
if (moe_intermediate_size := self.hparams.get("moe_intermediate_size")) is not None:
self.gguf_writer.add_expert_feed_forward_length(moe_intermediate_size)
if (shared_expert_intermediate_size := self.hparams.get('intermediate_size')) is not None and (num_key_value_heads := self.hparams.get('num_key_value_heads')) is not None:
self.gguf_writer.add_expert_shared_feed_forward_length(shared_expert_intermediate_size // num_key_value_heads)
if (shared_expert_count := self.hparams.get('moe_num_shared_experts')) is not None:
self.gguf_writer.add_expert_shared_count(shared_expert_count)
if shared_expert_count > 0 and (shared_expert_intermediate_size := self.hparams.get('intermediate_size')) is not None and (num_key_value_heads := self.hparams.get('num_key_value_heads')) is not None:
self.gguf_writer.add_expert_shared_feed_forward_length(shared_expert_intermediate_size // num_key_value_heads)
def modify_tensors(self, data_torch: Tensor, name: str, bid: int | None) -> Iterable[tuple[str, Tensor]]:
# Modify correction bias name as in DeepseekV2