Katehuuh

and also please add `preprocessing_num_workers` in WebUI.

Already said it up, Dataset: [pokemon_1k ](https://huggingface.co/datasets/BUAADreamer/pokemon-gpt4-1k): > set CUDA_VISIBLE_DEVICES=0 && llamafactory-cli train --stage sft --do_train True --model_name_or_path llava-hf/llava-1.5-13b-hf --preprocessing_num_workers 16 --finetuning_type lora --quantization_bit 8 --template vicuna --rope_scaling linear --flash_attn...

Tested using pair QA dataset, 20epoch `sampleDiffAppleOrange.json`: ``` [ { "instruction": "Which fruit is preferred by Katehuuh?", "input": "", "output": "Katehuuh prefers apples." }, { "instruction": "Can you tell me...

@hiyouga oi, mind checking PR? 😁

> There is a 60 second timeout that we'll wait for the next chunk of audio, otherwise there is a risk of deadlock. Well... probably given the long required output...

I've tried extending gradio/webrtc timeout but generator remains locked. Then tested `threading.Thread(target=process_responses)` with state locks to handle LLM/TTS off-main-thread. This avoids the 60s timeout by processing long requests in background...

I have the same issue with [oobabooga/text-generation-webui - OpenAI-compatible API endpoint](https://github.com/oobabooga/text-generation-webui/blob/main/docs/12%20-%20OpenAI%20API.md) `set OPENAI_API_BASE=http://127.0.0.1:5000/v1` using Qwen 2.5 Coder. test Include modify ollama endpoint or [azure-openai](https://docs.browser-use.com/customize/supported-models#azure-openai), it fails due to parameter incompatibility...

~~This win-fork worked for me: https://github.com/nan0bug00/text-generation-webui exl2+llama.cpp whl.~~ - [x] Duplicate of #6897 – Fixed in main

@Ghost4242 This issue now fixed in main branch: https://github.com/oobabooga/text-generation-webui/issues/6897#issuecomment-2966493588

We can make a workaround, but simpler and slower, for Windows visualization by modifying elastic_dragon.py --vis to render a GIF using matplotlib instead of OpenGL. 1. Add imports at top:...