Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

There is some problem with the LLaVA_13b_8bit_colab.ipynb #5

Open
xinsir6 opened this issue Nov 14, 2023 · 1 comment
Open

There is some problem with the LLaVA_13b_8bit_colab.ipynb #5

xinsir6 opened this issue Nov 14, 2023 · 1 comment

Comments

@xinsir6
Copy link

xinsir6 commented Nov 14, 2023

image
I use the ipynb scripts and load the model but always got Error:
image
The complete information is :
NotImplementedError Traceback (most recent call last)
Cell In[2], line 10
7 model_path = "/dfs/comicai/qi.xin/models/llava-v1.5-13b-5GB"
8 # model_path = "4bit/llava-v1.5-13b-4GB-8bit"
9 # model = LlavaLlamaForCausalLM.from_pretrained(model_path)
---> 10 model = LlavaLlamaForCausalLM.from_pretrained(model_path, low_cpu_mem_usage=True, load_in_8bit=True, device_map="auto")
12 kwargs = {"device_map": "auto"}
13 kwargs['load_in_4bit'] = True

File ~/anaconda3/envs/llava/lib/python3.10/site-packages/transformers/modeling_utils.py:2903, in PreTrainedModel.from_pretrained(cls, pretrained_model_name_or_path, config, cache_dir, ignore_mismatched_sizes, force_download, local_files_only, token, revision, use_safetensors, *model_args, **kwargs)
2893 if dtype_orig is not None:
2894 torch.set_default_dtype(dtype_orig)
2896 (
2897 model,
2898 missing_keys,
2899 unexpected_keys,
2900 mismatched_keys,
2901 offload_index,
2902 error_msgs,
-> 2903 ) = cls._load_pretrained_model(
2904 model,
2905 state_dict,
2906 loaded_state_dict_keys, # XXX: rename?
2907 resolved_archive_file,
2908 pretrained_model_name_or_path,
2909 ignore_mismatched_sizes=ignore_mismatched_sizes,
2910 sharded_metadata=sharded_metadata,
2911 _fast_init=_fast_init,
2912 low_cpu_mem_usage=low_cpu_mem_usage,
2913 device_map=device_map,
2914 offload_folder=offload_folder,
2915 offload_state_dict=offload_state_dict,
2916 dtype=torch_dtype,
2917 is_quantized=(load_in_8bit or load_in_4bit),
2918 keep_in_fp32_modules=keep_in_fp32_modules,
2919 )
2921 model.is_loaded_in_4bit = load_in_4bit
2922 model.is_loaded_in_8bit = load_in_8bit

File ~/anaconda3/envs/llava/lib/python3.10/site-packages/transformers/modeling_utils.py:3260, in PreTrainedModel._load_pretrained_model(cls, model, state_dict, loaded_keys, resolved_archive_file, pretrained_model_name_or_path, ignore_mismatched_sizes, sharded_metadata, _fast_init, low_cpu_mem_usage, device_map, offload_folder, offload_state_dict, dtype, is_quantized, keep_in_fp32_modules)
3250 mismatched_keys += _find_mismatched_keys(
3251 state_dict,
3252 model_state_dict,
(...)
3256 ignore_mismatched_sizes,
3257 )
3259 if low_cpu_mem_usage:
-> 3260 new_error_msgs, offload_index, state_dict_index = _load_state_dict_into_meta_model(
3261 model_to_load,
3262 state_dict,
3263 loaded_keys,
3264 start_prefix,
3265 expected_keys,
3266 device_map=device_map,
3267 offload_folder=offload_folder,
3268 offload_index=offload_index,
3269 state_dict_folder=state_dict_folder,
3270 state_dict_index=state_dict_index,
3271 dtype=dtype,
3272 is_quantized=is_quantized,
3273 is_safetensors=is_safetensors,
3274 keep_in_fp32_modules=keep_in_fp32_modules,
3275 )
3276 error_msgs += new_error_msgs
3277 else:

File ~/anaconda3/envs/llava/lib/python3.10/site-packages/transformers/modeling_utils.py:725, in _load_state_dict_into_meta_model(model, state_dict, loaded_state_dict_keys, start_prefix, expected_keys, device_map, offload_folder, offload_index, state_dict_folder, state_dict_index, dtype, is_quantized, is_safetensors, keep_in_fp32_modules)
722 fp16_statistics = None
724 if "SCB" not in param_name:
--> 725 set_module_quantized_tensor_to_device(
726 model, param_name, param_device, value=param, fp16_statistics=fp16_statistics
727 )
729 return error_msgs, offload_index, state_dict_index

File ~/anaconda3/envs/llava/lib/python3.10/site-packages/transformers/utils/bitsandbytes.py:77, in set_module_quantized_tensor_to_device(module, tensor_name, device, value, fp16_statistics)
75 new_value = old_value.to(device)
76 elif isinstance(value, torch.Tensor):
---> 77 new_value = value.to("cpu")
78 if value.dtype == torch.int8:
79 is_8bit_serializable = version.parse(importlib.metadata.version("bitsandbytes")) > version.parse(
80 "0.37.2"
81 )

NotImplementedError: Cannot copy out of meta tensor; no data!

So it there some problem the model in huggingface?

@steinhaug
Copy link

I'm quite sure there is a problem with the model from huggingface, I get the same problems trying to load it.

llava-v1.5-7b and llava-v1.5-13b-3GB works fine.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

2 participants