From 8030718fa8e796b9567035bfbfe088666cff4757 Mon Sep 17 00:00:00 2001 From: zhangfaen Date: Fri, 6 Sep 2024 16:37:38 +0800 Subject: [PATCH 1/2] add an alternative to LlamaFactory --- README.md | 2 ++ 1 file changed, 2 insertions(+) diff --git a/README.md b/README.md index 015faaa..9a89828 100644 --- a/README.md +++ b/README.md @@ -1368,6 +1368,8 @@ hyperparameters. One argument to note is ``cutoff_len``, which is the maximum length of the training data. Control this parameter to avoid OOM error. +For finetuning Qwen2-VL, in case you prefer a lightweight way, e.g. you would like to write your own "train loop". Here is an alternative repo for reference: [Qwen2-VL-Finetuning](https://github.com/zhangfaen/finetune-Qwen2-VL) . The are about 200 lines of effective code for preparing dataset, dataloader, train loop, etc. It is simple and straight forward. + ## Function Calling Qwen2-VL supports Function Calling (aka. Tool Calling or Tool Use). For details on how to use this capability, please refer to the Qwen-Agent project for [the function calling example](https://github.com/QwenLM/Qwen-Agent/blob/main/examples/qwen2vl_function_calling.py) and [the agent example](https://github.com/QwenLM/Qwen-Agent/blob/main/examples/qwen2vl_assistant_tooluse.py). From 42ae1670055a5202495eda368a9f1d948b254ca3 Mon Sep 17 00:00:00 2001 From: zhangfaen Date: Fri, 13 Sep 2024 13:09:07 +0800 Subject: [PATCH 2/2] refine wording --- README.md | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/README.md b/README.md index 9a89828..94aaf71 100644 --- a/README.md +++ b/README.md @@ -1368,7 +1368,7 @@ hyperparameters. One argument to note is ``cutoff_len``, which is the maximum length of the training data. Control this parameter to avoid OOM error. -For finetuning Qwen2-VL, in case you prefer a lightweight way, e.g. you would like to write your own "train loop". Here is an alternative repo for reference: [Qwen2-VL-Finetuning](https://github.com/zhangfaen/finetune-Qwen2-VL) . The are about 200 lines of effective code for preparing dataset, dataloader, train loop, etc. It is simple and straight forward. +**Note:** For finetuning Qwen2-VL, in case you prefer a lightweight way, e.g. you would like to write your own "train loop". Here is an alternative repo for reference: [Qwen2-VL-Finetuning](https://github.com/zhangfaen/finetune-Qwen2-VL) . There are about 200 lines of effective code for preparing dataset, dataloader, train loop, etc. It is simple and straight forward. It supports flash_attention2 and distributed training. ## Function Calling