Update workflow.py

[renjie130]

User description

What does this PR do?

Fixes # (issue)

Before submitting

• Did you read the contributor guideline?
• Did you write any new necessary tests?

---

PR Type

Enhancement

---

Description

• 引入了 DataCollatorForLanguageModeling 以简化数据整理过程。
• 将 CustomSeq2SeqTrainer 替换为 CustomTrainer，可能改进了训练逻辑。
• 移除了 SFTDataCollatorWith4DAttentionMask，简化了代码结构。

---

Changes walkthrough 📝

	Relevant files
增强	workflow.py 更新训练工作流                                                                                                    src/llamafactory/train/sft/workflow.py 引入了 DataCollatorForLanguageModeling 用于数据整理。 将 CustomSeq2SeqTrainer 替换为 CustomTrainer。 移除了 SFTDataCollatorWith4DAttentionMask 的使用。 +4/-12

---

💡 PR-Agent usage: Comment /help "your question" on any pull request to receive relevant information

[renjie130]

PR Reviewer Guide 🔍

(Review updated until commit 8e9f6b8)

Here are some key observations to aid the review process:

⏱️ Estimated effort to review: 2 🔵🔵⚪⚪⚪

🏅 Score: 85

🧪 PR contains tests

🔒 Security concerns 否

⚡ Recommended focus areas for review 潜在功能影响 将 CustomSeq2SeqTrainer 替换为 CustomTrainer 可能会影响训练逻辑，需要验证是否所有功能都得到保留。 trainer = CustomTrainer( model=model, args=training_args, finetuning_args=finetuning_args, data_collator=data_collator, callbacks=callbacks, **dataset_module, **tokenizer_module, ) 数据整理器变更 使用 DataCollatorForLanguageModeling 替换 SFTDataCollatorWith4DAttentionMask 可能会影响数据处理逻辑，需要验证是否所有功能都得到保留。 data_collator = DataCollatorForLanguageModeling(tokenizer=tokenizer, mlm=False)

[renjie130]

PR Description updated to latest commit (8e9f6b8)

[renjie130]

PR Code Suggestions ✨

Latest suggestions up to 8e9f6b8

Category	Suggestion	Score
关键错误	确保数据整理器与训练器的需求兼容。 确保 data_collator 的配置与 CustomTrainer 的需求一致，因为 DataCollatorForLanguageModeling 可能不支持某些特定的训练需求。 src/llamafactory/train/sft/workflow.py [59] -data_collator = DataCollatorForLanguageModeling(tokenizer=tokenizer, mlm=False) +data_collator = SFTDataCollatorWith4DAttentionMask(...) Suggestion importance[1-10]: 8 Why: 该建议指出了潜在的关键兼容性问题，确保数据整理器与训练器的需求一致非常重要。	8

---

Previous suggestions

Suggestions up to commit 8e9f6b8

Category	Suggestion	Score
关键错误	确保数据整理器与训练器的需求一致。 确保 data_collator 的配置与 CustomTrainer 的需求一致，因为 DataCollatorForLanguageModeling 可能不包含 SFTDataCollatorWith4DAttentionMask 的所有功能。 src/llamafactory/train/sft/workflow.py [59] -data_collator = DataCollatorForLanguageModeling(tokenizer=tokenizer, mlm=False) +data_collator = SFTDataCollatorWith4DAttentionMask( + template=template, + model=model if not training_args.predict_with_generate else None, + pad_to_multiple_of=8 if training_args.do_train else None, + label_pad_token_id=IGNORE_INDEX if data_args.ignore_pad_token_for_loss else tokenizer.pad_token_id, + block_diag_attn=model_args.block_diag_attn, + attn_implementation=getattr(model.config, "_attn_implementation", None), + compute_dtype=model_args.compute_dtype, + **tokenizer_module, +) Suggestion importance[1-10]: 10 Why: 这是一个关键错误，因为 DataCollatorForLanguageModeling 可能不包含 SFTDataCollatorWith4DAttentionMask 的所有功能，可能导致训练失败或结果不准确。	10

Suggestions up to commit 8e9f6b8

Category	Suggestion	Score
关键错误	恢复使用 SFTDataCollatorWith4DAttentionMask 以确保数据整理器与模型和训练参数兼容。 确保 data_collator 使用 SFTDataCollatorWith4DAttentionMask 而不是 DataCollatorForLanguageModeling，以保持与模型和训练参数的兼容性。 src/llamafactory/train/sft/workflow.py [59] -data_collator = DataCollatorForLanguageModeling(tokenizer=tokenizer, mlm=False) +data_collator = SFTDataCollatorWith4DAttentionMask( + template=template, + model=model if not training_args.predict_with_generate else None, + pad_to_multiple_of=8 if training_args.do_train else None, + label_pad_token_id=IGNORE_INDEX if data_args.ignore_pad_token_for_loss else tokenizer.pad_token_id, + block_diag_attn=model_args.block_diag_attn, + attn_implementation=getattr(model.config, "_attn_implementation", None), + compute_dtype=model_args.compute_dtype, + **tokenizer_module, +) Suggestion importance[1-10]: 10 Why: 该建议解决了关键错误，确保数据整理器与模型和训练参数的兼容性，避免潜在的功能问题。	10

Suggestions up to commit 8e9f6b8

Category	Suggestion	Score
关键错误	确保数据整理器的配置与训练器的需求一致。 确保 data_collator 的配置与 CustomTrainer 的需求一致，避免潜在的训练或评估错误。 src/llamafactory/train/sft/workflow.py [59] -data_collator = DataCollatorForLanguageModeling(tokenizer=tokenizer, mlm=False) +data_collator = SFTDataCollatorWith4DAttentionMask(...) Suggestion importance[1-10]: 10 Why: 该建议指出了关键错误，data_collator 的配置与 CustomTrainer 的需求不一致，可能导致训练或评估错误。	10

[renjie130]

Persistent review updated to latest commit 8e9f6b8

[renjie130]

Persistent review updated to latest commit 8e9f6b8

[renjie130]

Failed to generate code suggestions for PR