-
I think that the GQA in CodeLlama-34B is messing with the Flash Attention Monkey Patch.
When training with the monkey patch, I get errors like:
> File "/fsx/training/llama/flash_attn_monkey_pat…
-
Hi!
Thank you for this great work.
I'm trying to run [SDTurbo](https://huggingface.co/stabilityai/sd-turbo) with diffusers.js.
I've followed the instructions from [this issue](https://github.…
-
您好,llama_attn_replace_sft.py中forward_noflashattn()方法中有加一行关于q_len和group_size关系的判断,如下:
def forward_noflashattn(
self,
hidden_states: torch.Tensor,
attention_mask: Optional[torch.Tensor…
-
-
### 描述该错误
在internlm-7B基础上用自己的sft数据训练,并转化为hf格式,加载模型进行推理,结果出现乱码
![image](https://github.com/InternLM/InternLM/assets/44628671/470ab99a-ae95-48df-b706-54d456237955)
### 环境信息
```
response, history …
-
In model.py, the `toekn` is pasted to `self.tok_embeddings = Linear(params.vocab_size, params.dim)` in `forward()` function. But in generation.py, `token` was defined as `tokens = torch.full((bsz, tot…
-
Da wir standardmäßig nach der Übertragung der Daten aus OPUS in die Verbund-Datenbank die Verbund-ID zurückspielen, würden wir dies gerne in ein Standard-Identifier-Feld speichern.
Werte sind aktue…
-
Hello! I'm getting an error when running the `example_chat_completion.py` script. Any help is much appreciated. Thank you!
```
torchrun --nproc_per_node 1 example_chat_completion.py \
--ckpt_…
-
I am using 4 3090ti cards, and I have set the batch size to very small, but this situation occurs every time the first epoch is clicked
Traceback (most recent call last):
File "train.py", line…
-
**Describe the bug**
SD.Next recently switched to an internal LoRA/LyCORIS handler. When using Regional Prompter in *Latent* mode, the following traceback is raised.
```
12:44:57-273233 ERROR …