[assets] update readme (#7644)

README.md

@@ -42,6 +42,7 @@ Choose your path:
 - **Local machine**: Please refer to [usage](#getting-started)
 - **PAI-DSW (free trial)**: [Llama3 Example](https://gallery.pai-ml.com/#/preview/deepLearning/nlp/llama_factory) | [Qwen2-VL Example](https://gallery.pai-ml.com/#/preview/deepLearning/nlp/llama_factory_qwen2vl) | [DeepSeek-R1-Distill Example](https://gallery.pai-ml.com/#/preview/deepLearning/nlp/llama_factory_deepseek_r1_distill_7b)
 - **Amazon SageMaker**: [Blog](https://aws.amazon.com/cn/blogs/china/a-one-stop-code-free-model-fine-tuning-deployment-platform-based-on-sagemaker-and-llama-factory/)
+- **Easy Dataset**: [Fine-tune on Synthetic Data](https://buaa-act.feishu.cn/wiki/GVzlwYcRFiR8OLkHbL6cQpYin7g)
 
 > [!NOTE]
 > Except for the above links, all other websites are unauthorized third-party websites. Please carefully use them.
@@ -557,6 +558,8 @@ Please refer to [data/README.md](data/README.md) for checking the details about
 > [!NOTE]
 > Please update `data/dataset_info.json` to use your custom dataset.
 
+You can also use **[Easy Dataset](https://github.com/ConardLi/easy-dataset)** to create synthetic data for fine-tuning.
+
 ### Quickstart
 
 Use the following 3 commands to run LoRA **fine-tuning**, **inference** and **merging** of the Llama3-8B-Instruct model, respectively.

README_zh.md

@@ -44,6 +44,7 @@ https://github.com/user-attachments/assets/43b700c6-a178-41db-b1f8-8190a5d3fcfc
 - **本地机器**：请见[如何使用](#如何使用)
 - **PAI-DSW（免费试用）**：[Llama3 案例](https://gallery.pai-ml.com/#/preview/deepLearning/nlp/llama_factory) | [Qwen2-VL 案例](https://gallery.pai-ml.com/#/preview/deepLearning/nlp/llama_factory_qwen2vl) | [DeepSeek-R1-Distill 案例](https://gallery.pai-ml.com/#/preview/deepLearning/nlp/llama_factory_deepseek_r1_distill_7b)
 - **Amazon SageMaker**：[博客](https://aws.amazon.com/cn/blogs/china/a-one-stop-code-free-model-fine-tuning-deployment-platform-based-on-sagemaker-and-llama-factory/)
+- **Easy Dataset**：[数据蒸馏微调](https://buaa-act.feishu.cn/wiki/KY9xwTGs1iqHrRkjXBwcZP9WnL9)
 
 > [!NOTE]
 > 除上述链接以外的其他网站均为未经许可的第三方网站，请小心甄别。
@@ -404,7 +405,7 @@ huggingface-cli login
 
 ## 软硬件依赖
 
-| 必需项       | 至少     | 推荐      |
+| 必需项        | 至少     | 推荐      |
 | ------------ | ------- | --------- |
 | python       | 3.9     | 3.10      |
 | torch        | 1.13.1  | 2.6.0     |
@@ -414,7 +415,7 @@ huggingface-cli login
 | peft         | 0.14.0  | 0.15.0    |
 | trl          | 0.8.6   | 0.9.6     |
 
-| 可选项       | 至少     | 推荐      |
+| 可选项        | 至少     | 推荐      |
 | ------------ | ------- | --------- |
 | CUDA         | 11.6    | 12.2      |
 | deepspeed    | 0.10.0  | 0.16.4    |
@@ -560,6 +561,8 @@ pip install .
 > [!NOTE]
 > 使用自定义数据集时，请更新 `data/dataset_info.json` 文件。
 
+您也可以使用 **[Easy Dataset](https://github.com/ConardLi/easy-dataset)** 构建用于微调的合成数据。
+
 ### 快速开始
 
 下面三行命令分别对 Llama3-8B-Instruct 模型进行 LoRA **微调**、**推理**和**合并**。

data/mllm_video_audio_demo.json

@@ -53,5 +53,4 @@
       "mllm_demo_data/4.mp3"
     ]
   }
-  
 ]

requirements.txt

@@ -1,4 +1,4 @@
-transformers>=4.41.2,<=4.51.0,!=4.46.*,!=4.47.*,!=4.48.0
+transformers>=4.41.2,<=4.51.1,!=4.46.*,!=4.47.*,!=4.48.0
 datasets>=2.16.0,<=3.4.1
 accelerate>=0.34.0,<=1.5.2
 peft>=0.14.0,<=0.15.0

setup.py

@@ -54,7 +54,7 @@ extra_require = {
     "awq": ["autoawq"],
     "aqlm": ["aqlm[gpu]>=1.1.0"],
     "vllm": ["vllm>=0.4.3,<=0.8.3"],
-    "sglang": ["sglang[srt]>=0.4.5", "transformers==4.51.0"],
+    "sglang": ["sglang[srt]>=0.4.5", "transformers==4.51.1"],
     "galore": ["galore-torch"],
     "apollo": ["apollo-torch"],
     "badam": ["badam>=1.2.1"],

src/llamafactory/__init__.py

@@ -19,7 +19,7 @@ Level:
 
 Dependency graph:
   main:
-    transformers>=4.41.2,<=4.51.0,!=4.46.*,!=4.47.*,!=4.48.0
+    transformers>=4.41.2,<=4.51.1,!=4.46.*,!=4.47.*,!=4.48.0
     datasets>=2.16.0,<=3.4.1
     accelerate>=0.34.0,<=1.5.2
     peft>=0.14.0,<=0.15.0

src/llamafactory/data/collator.py

@@ -298,8 +298,9 @@ class KTODataCollatorWithPadding(MultiModalDataCollatorForSeq2Seq):
         batch["kl_input_ids"] = kl_batch["input_ids"]
         batch["kl_attention_mask"] = kl_batch["attention_mask"]
         batch["kl_labels"] = kl_batch["labels"]
-        if "cross_attention_mask" in kl_batch:  # for mllama inputs.
+        if "cross_attention_mask" in kl_batch:  # for mllama inputs
             batch["kl_cross_attention_mask"] = kl_batch["cross_attention_mask"]
+
         if "token_type_ids" in kl_batch:
             batch["kl_token_type_ids"] = kl_batch["token_type_ids"]
 

src/llamafactory/extras/misc.py

@@ -89,7 +89,7 @@ def check_version(requirement: str, mandatory: bool = False) -> None:
 
 def check_dependencies() -> None:
     r"""Check the version of the required packages."""
-    check_version("transformers>=4.41.2,<=4.51.0,!=4.46.0,!=4.46.1,!=4.46.2,!=4.46.3,!=4.47.0,!=4.47.1,!=4.48.0")
+    check_version("transformers>=4.41.2,<=4.51.1,!=4.46.0,!=4.46.1,!=4.46.2,!=4.46.3,!=4.47.0,!=4.47.1,!=4.48.0")
     check_version("datasets>=2.16.0,<=3.4.1")
     check_version("accelerate>=0.34.0,<=1.5.2")
     check_version("peft>=0.14.0,<=0.15.0")

src/llamafactory/train/kto/trainer.py

@@ -147,6 +147,9 @@ class CustomKTOTrainer(KTOTrainer):
         if "pixel_values" in batch:
             model_inputs["pixel_values"] = batch["pixel_values"]
 
+        if "image_sizes" in batch:
+            model_inputs["image_sizes"] = batch["image_sizes"]
+
         if "image_grid_thw" in batch:
             model_inputs["image_grid_thw"] = batch["image_grid_thw"]