423A35C7/LLaMA-Factory
1
0
Fork 0
mirror of https://github.com/hiyouga/LLaMA-Factory.git synced 2025-10-14 23:58:11 +08:00
Code Issues Packages Projects Releases Wiki Activity

[data] Fix wrong position ids with packed attention masks (#7754)

Browse Source 
Co-authored-by: hoshi-hiyouga <hiyouga@buaa.edu.cn>
This commit is contained in:
Changrui Chen 2025-04-21 16:19:36 +01:00 committed by GitHub
parent 0ac641326b
commit bd7bc31c79
1 changed files with 1 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
src/llamafactory/data/collator.py
View File

@ -176,7 +176,7 @@ class MultiModalDataCollatorForSeq2Seq(DataCollatorForSeq2Seq):
"input_ids": features["input_ids"], "input_ids": features["input_ids"],
"image_grid_thw": mm_inputs.get("image_grid_thw"), "image_grid_thw": mm_inputs.get("image_grid_thw"),
"video_grid_thw": mm_inputs.get("video_grid_thw"), "video_grid_thw": mm_inputs.get("video_grid_thw"),
"attention_mask": features["attention_mask"], "attention_mask": (features["attention_mask"] >= 1).float(),
} }
if "second_per_grid_ts" in mm_inputs: # for qwen2vl if "second_per_grid_ts" in mm_inputs: # for qwen2vl
rope_index_kwargs["second_per_grid_ts"] = mm_inputs.get("second_per_grid_ts") rope_index_kwargs["second_per_grid_ts"] = mm_inputs.get("second_per_grid_ts")