Yi

Yi copied to clipboard

Yi-34B微调后，中文的问题，突然用英文回答，回答的虽然对，但是语言没对齐

1

同题目

QingqingSun-Bao

sft报错：ValueError: YiForCausalLM does not support Flash Attention 2.0 yet.

8

根据README运行sft脚本： ``` cd finetune/scripts bash run_sft_Yi_6b.sh ``` **报错信息** ``` [2024-01-02 10:43:01,920] [INFO] [real_accelerator.py:158:get_accelerator] Setting ds_accelerator to cuda (auto detect) [2024-01-02 10:43:04,373] [WARNING] [runner.py:203:fetch_hostfile] Unable to find hostfile, will proceed with...

zhangxiann

用 VLLM 加载 Yi-34B-Chat-4bits-GPTQ, 模型有时不停地输出空字串而不停止

7

``` sampling params: SamplingParams(n=1, best_of=1, presence_penalty=0.0, frequency_penalty=0.0, repetition_penalty=1.0, temperature=0.7, top_p=1.0, top_k=-1, min_p=0.0, use_beam_search=False, length_penalty=1.0, early_stopping=False, stop=[], stop_token_ids=[7], include_stop_str_in_output=False, ignore_eos=False, max_tokens=3736, logprobs=None, prompt_logprobs=None, skip_special_tokens=True, spaces_between_special_tokens=True) ``` 在v100 32GB上运行，启动脚本为 ```sh CUDA_VISIBLE_DEVICES=2,3 python...

zhanghx0905

微调 YI-6B 一直出现 loss scale overflow 然后 reduce 到 min_loss_scale 报错, YI-6B-Chat 则没问题，chat 模型训练参数设置有什么不同吗

4

  环境： V100 * 8 deepspeed: 0.10.3 torch: 2.1.0+cu118

xinbingzhe

您好，请问Yi是否通过了大模型备案。

LiuChen19960902

RuntimeError: You can't move a model that has some modules offloaded to cpu or disk.

5

You shouldn't move a model when it is dispatched on multiple devices. Traceback (most recent call last): File "/home/yi/workspace/Yi/finetune/sft/main.py", line 416, in main() File "/home/yi/workspace/Yi/finetune/sft/main.py", line 345, in main model,...

dhcn

Yi-34B-Chat-4bit 无法配合langchain作为agent使用

2

我使用34B-chat和langchain 配合作为一个agent使用。 prompt如下： ``` ou are a friendly Chatbot that having a conversation with a human. You NEVER answer that you are an AI language model. If you don't know...

weiminw

Function Calling 功能支持计划

1

请问后期是否会发布加入Function Calling微调后的版本？

nemoiee

就不能在官网挂一个体验的链接吗

1

就不能在官网挂一个体验的链接吗？找了半天，官网，公众号，开源社区等等，连一个体验链接都找不到。你们官方部署一个体验的环境没那么难吗？ 我这边正好有客户需要处理长上下文本的需求，总不能每个要体验的人都自己去部署吧

dfengpo

## Bug Description In instruct mode, the model do not stop after completing the first valid assistant response (). It then generates garbage, even "user"... ``` system system 你是一个AI助手。 user...

dingmaotu