Open xiaotuoluo11 opened 1 year ago
运行指令: deepspeed --master_port=24999 train_ds.py --version='./llava_path' --dataset_dir='./dataset' --vision_pretrained='sam_weights' --dataset='ade20k' --sample_rates="9,3,3,1" --exp_name="lisa-7b" --load_in_8bit
下面是llava_path 下的文件
config.json mm_projector.bin pytorch_model-00002-of-00003.bin pytorch_model.bin.index.json tokenizer_config.json
generation_config.json pytorch_model-00001-of-00003.bin pytorch_model-00003-of-00003.bin special_tokens_map.json tokenizer.model
使用了上述指令和权重文件报错:
You are using the legacy behaviour of the <class 'transformers.models.llama.tokenization_llama.LlamaTokenizer'>. This means that tokens that come after special tokens will not be properly handled. We recommend you to read the related pull request available at https://github.com/huggingface/transformers/pull/24565
normalizer.cc(51) LOG(INFO) precompiled_charsmap is empty. use identity normalization.
You are using a model of type llama to instantiate a model of type llava. This is not supported for all configurations of models and can yield errors.
Traceback (most recent call last):
File "/home/user/project/detgpt/LISA/train_ds.py", line 461, in
应该是llava没有下载或者转换成功,建议检查一下。另外,训练不能使用--load_in_8bit
命令。
另外我们重构了新版的代码,欢迎使用!
llava权重按照说明下载完了,并且在上面进行了列举,您看看缺少那些文件
Hi @X-Lai @xiaotuoluo11 , did you solve this issue - I have been trying to train the model on cloud, and I get this error :
Traceback (most recent call last):
File "/opt/ml/code/train_ds.py", line 593, in repo_type
argument if needed.
My train script is this: deepspeed --master_port=24999 /opt/ml/code/train_ds.py \ --version="/tmp/LISA_ft/downloads" \ --dataset_dir='/tmp/LISA_ft/dataset' \ --vision_pretrained="/opt/ml/code/sam_vit_h_4b8939.pth" \ --dataset="sem_seg||refer_seg||vqa||reason_seg" \ --sample_rates="9,3,3,1" \ --exp_name="lisa-13b" \ --batch_size="2" \
I tried setting the TRANSFORMERS_CACHE to /tmp/LISA_ft directory as well. However, it gave a different error: OSError: /tmp/LISA_ft/downloads does not appear to have a file named config.json. Checkout 'https://huggingface.co//tmp/LISA_ft/downloads/None' for available files.
Can you please help identify the issue?
运行指令: deepspeed --master_port=24999 train_ds.py --version='./llava_path' --dataset_dir='./dataset' --vision_pretrained='sam_weights' --dataset='ade20k' --sample_rates="9,3,3,1" --exp_name="lisa-7b" --load_in_8bit