423A35C7/LLaMA-Factory
1
0
Fork 0
mirror of https://github.com/hiyouga/LLaMA-Factory.git synced 2026-05-28 19:08:57 +08:00
Code Issues Packages Projects Releases Wiki Activity

[train] add qwen35 patch for neat_packing (#10436)

Browse Source
This commit is contained in:
Kingsley
2026-04-27 00:31:49 +08:00
committed by GitHub
parent e0bc3c1971
commit 79c8332e4c
2 changed files with 190 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
src/llamafactory/data/collator.py
View File

@@ -471,8 +471,8 @@ class SFTDataCollatorWith4DAttentionMask(MultiModalDataCollatorForSeq2Seq):
def __post_init__(self):
super().__post_init__()
if self.neat_packing and self.attn_implementation == "flash_attention_2":
if self.model is not None and getattr(self.model.config, "model_type", None) in ["qwen3_5", "qwen3_5_moe", "gpt_oss"]:
raise ValueError("Neat packing is not supported for qwen3_5, qwen3_5_moe, gpt_oss models for now.")
if self.model is not None and getattr(self.model.config, "model_type", None) in ["gemma4", "gpt_oss"]:
raise ValueError("Neat packing is not supported for gemma4, gpt_oss models for now.")
@staticmethod
def _unpad_packed_features(features: dict[str, Any]) -> None: