Xinlin Zhuang comments

Results 7 comments of


                                            Xinlin Zhuang

Need help for installing dolma

Here is the Rust version: ``` rustc 1.78.0 (9b00956e5 2024-04-29) ```

Need help for installing dolma

Hi, thanks for your suggestion! It appears that my CMake version is too old. This problem has been solved. :)

step 2 when run "/get_train_lora_grads.sh", load the optimizer.pt error is happend

> @tengerye I solve the error by adding `--fsdp 'full_shard auto_wrap' --fsdp_config llama_finetune` to `less/scripts/train/warmup_lora_train.sh`. The code works well. > > Here is the changed version. > > ```shell >...

[Feature] Please support Llama3.2 and Qwen2.5

> 就模型结构而言，qwen2.5==qwen2，我已经跑起来了 Hi，但是我在用TurboMind Qwen 2.5会报错，提示不支持

[Feature] Please support Llama3.2 and Qwen2.5

> pytorch是可以的，turbomind或许需要适配，你可以尝试直接改代码底层，看能不能让引擎将qwen2.5认成qwen2 谢谢您的建议！在官方正式支持前，也只能这么干了

[Usage]: What's the relationship between KV cache and MAX_SEQUENCE_LENGTH.

Same question.

Does every sample get the same demonstrations through vote_k retriver?

According to my understanding (Alg 1, Page 21, from the paper https://openreview.net/forum?id=qY1hlv7gwg), the current version only implements the step 1, which construts the candidate set for further selection. It should...