LMFlow

LMFlow copied to clipboard

Does it or will it support Vicuna-13b-v1.1 finetuning?

1

Thanks for creating such an amazing project. Vicuna-13b-v1.1 seems to achieve impressive performance in tasks in multiple languages. I believe it would be a good idea to conduct further fine-tuning...

ghost

Hardware spec for finetuning >7B Llama

22

Hi, thanks for this nice repo! I'm always facing kill signal and error code -11 after executing example/finetune.py with my own 10k dataset (text_only) on a single A100 40GB, CPU...

ChaoChungWu-Johnson

Is there any restrictions on the length of the model output?

1

I ran the ./scripts/run_evaluation_with_lora.sh with my testing data. However, the string length of the model output seems to have a limit (Some strings in the testing data is very long),...

szl666

Is it necessary to use Ampere GPU?

2

运行环境是colab，gpu是tsla T4，torch版本是2.0，CUDA版本是11.8，训练模型的时候出现这个报错：ValueError: Your setup doesn't support bf16/gpu. You need torch>=1.10, using Ampere GPU with cuda>=11.0

GitHubJKin

no error log and exits with return code = -7

3

+ deepspeed --master_port=11000 examples/finetune.py --model_name_or_path /workspace/work/LMFlow/LMFlow/mydata/model/llama-7b-hf --save_aggregated_lora 0 --use_lora 1 --lora_r 8 --dataset_path /workspace/work/LMFlow/LMFlow/mydata/data/wiki_cn --block_size 512 --validation_split_percentage 0 --dataloader_num_workers 1 --output_dir /workspace/work/LMFlow/LMFlow/mydata/model/output_models/7b-wiki --overwrite_output_dir --num_train_epochs 0.01 --learning_rate 1e-4 --per_device_train_batch_size 4 --deepspeed...

delltower

launch.py:sigkill_handler exits with return code = -11

1

我以为是设备资源的问题，因此也尝试将模型换成`git-2 samll`，也出现相同的错误。 这个模型是是来自`https://huggingface.co/models?pipeline_tag=text-generation&sort=downloads&search=gpt2-small-chinese-cluecorpussmall`的`gpt2-small-chinese-cluecorpussmall`。 我怀疑问题可能并不在于设备的资源上。 ``` (lmflow) [root@a4113ca43b08 LMFlow-main]# ./scripts/run_finetune.sh [2023-04-15 16:04:25,114] [WARNING] [runner.py:186:fetch_hostfile] Unable to find hostfile, will proceed with training with local resources only. [2023-04-15 16:04:25,127] [INFO] [runner.py:550:main] cmd...

zeroandexe

about customize

4

Can I customize the indicators related to my data set in this file? #!/bin/bash CUDA_VISIBLE_DEVICES=0 \ deepspeed examples/evaluate.py \ --answer_type medmcqa \ --model_name_or_path /root/autodl-tmp/model/BEELE/ \ --dataset_path /root/autodl-tmp/LMFlow/data/alpaca/test \ --deepspeed examples/ds_config.json...

w-JiqQian

Why is the lora_model so large after fine-tuning?

1

Why is the lora_model so large after fine-tuning? 

w-JiqQian

about deepspeed

1

I have question about deepspeed. Assuming that we did not use DeepSpeed in our code initially, if we now want to use it in our fine-tuning process, do we just...

chris111110

Do evaluation during training

1

When I read the parameters to be set to finetune.py, I am a little confused. since there are several parameters about evaluation during training: -- validation_file: I did not find...

Daniel-1997