JingKun Hao

为嘛微调之后2不如1好呢？

2

### Is your feature request related to a problem? Please describe. _No response_ ### Solutions 为嘛微调之后2不如1好呢？ ### Additional context _No response_

### Is your feature request related to a problem? Please describe. ChatGLM2-6b lora微调，train_loss一致下降，但eval_loss不降反增； 同样的数据，同样的配置参数，ChatGLM-6b，这两个loss都是下降的，咋回事呢？ ### Solutions eval_loss不降反增 ### Additional context eval_loss不降反增

does is support model from vllm？

1

### 💻 系统环境 | Operating System Windows ### 🌐 浏览器 | Browser Chrome ### 🐛 问题描述 | Bug Description 1 ### 🚦 期望结果 | Expected Behavior 1 ### 📷 复现步骤...

Support LLMs from vllm

1

Add support for loading LLM models from vllm package: https://github.com/vllm-project/vllm

httpx.ConnectError: [Errno 101] Network is unreachable

3

### 🐛 Describe the bug httpx.ConnectError: [Errno 101] Network is unreachable

'TTS' object has no attribute 'is_multi_speaker'

7

### Describe the bug Traceback (most recent call last): File "/home/house365ai/xxm/TTS/demo.py", line 16, in wav = tts.tts(text="Hello world!", speaker_wav="/home/house365ai/xxm/TTS/output/audio.wav", language="en") File "/home/house365ai/xxm/TTS/TTS/api.py", line 334, in tts self._check_arguments( File "/home/house365ai/xxm/TTS/TTS/api.py", line...

中文能力不咋地啊

3

### Tested versions 中文能力不咋地啊 ### System information 中文能力不咋地啊 ### Issue description 中文能力不咋地啊 ### Minimal reproduction example (MRE) 中文能力不咋地啊