JamePeng

`if torch.cuda.is_available():` ` torch.cuda.empty_cache()` ` torch.cuda.ipc_collect()` ` st.session_state["state"] = []`

Update MiniCPM-Llama3-V-2_5 streamlit demo

Done！ This update, based on the May 25, 2024 version of modeling_minicpmv.py, includes the following enhancements: 1. Introduction of repetition_penalty and top_k parameters to the st.sidebar, enabling users to adjust...

error image:  error dataset json data: `{ "id": "8887_zh", "image": "./train_val_images/train_images/3c92ab3c46017e79.jpg", "conversations": [ { "role": "user", "content": "\n请描述这张图片。" }, { "role": "assistant", "content": "描述: \n图片显示了指示合并和车道关闭的路标，道路上有交通锥和警用摩托车，暗示着道路施工或交通情况。\n\n文本：\n左侧：\"前方左车道封闭\"\n中心附近：一个橙色菱形标志，有两个黑色箭头指向对方，象征车道合并，无文本。\n右侧：一个橙色菱形标志，带有两个车道合并的符号和文字，“前方左车道封闭”。" }, { "role": "user",...

{'loss': 1.034, 'grad_norm': 6.085077285766602, 'learning_rate': 5e-07, 'epoch': 0.19} {'loss': 1.1023, 'grad_norm': 3.602856397628784, 'learning_rate': 5e-07, 'epoch': 0.19} 61%|███████████████████████████████████████████████████████████████████████████████████████████████████████████████ | 4881/8000 [3:19:00

So， what is the reason of this bug? I saw my dataset is fine.

same problem i meet

My fork project has added some updates of llama4: https://github.com/JamePeng/llama-cpp-python

Yes, maybe some conversion work is needed for the BF16 type to make the model work better.

You can try my fork code, https://github.com/JamePeng/llama-cpp-python/commit/8d981f0455b3adabf46417e3f9304ca6b70357ed this commit had fixed this problem!