Pickling error - cannot load on transformers==4.37.0.dev0
Great work again on TInyLlama!
I tried loading this on transformers==4.37.0.dev0 and sadly it doesn't work :( The Chat one successfully loads though!
config.json: 100%
560/560 [00:00<00:00, 36.3kB/s]
pytorch_model.bin: 100%
4.40G/4.40G [00:18<00:00, 239MB/s]
---------------------------------------------------------------------------
UnpicklingError Traceback (most recent call last)
/usr/local/lib/python3.10/dist-packages/transformers/modeling_utils.py in load_state_dict(checkpoint_file)
518
--> 519 return torch.load(checkpoint_file, map_location=map_location, weights_only=True)
520 except Exception as e:
6 frames
UnpicklingError: Weights only load failed. Re-running `torch.load` with `weights_only` set to `False` will likely succeed, but it can result in arbitrary code execution.Do it only if you get the file from a trusted source. WeightsUnpickler error: Unsupported operand 149
During handling of the above exception, another exception occurred:
UnicodeDecodeError Traceback (most recent call last)
UnicodeDecodeError: 'utf-8' codec can't decode byte 0xc1 in position 70: invalid start byte
During handling of the above exception, another exception occurred:
OSError Traceback (most recent call last)
/usr/local/lib/python3.10/dist-packages/transformers/modeling_utils.py in load_state_dict(checkpoint_file)
533 ) from e
534 except (UnicodeDecodeError, ValueError):
--> 535 raise OSError(
536 f"Unable to load weights from pytorch checkpoint file for '{checkpoint_file}' "
537 f"at '{checkpoint_file}'. "
OSError: Unable to load weights from pytorch checkpoint file for '/root/.cache/huggingface/hub/models--TinyLlama--TinyLlama-1.1B-intermediate-step-1431k-3T/snapshots/4b8dd7e43ec08c24ccaf89cbf67898cff53c95ae/pytorch_model.bin' at '/root/.cache/huggingface/hub/models--TinyLlama--TinyLlama-1.1B-intermediate-step-1431k-3T/snapshots/4b8dd7e43ec08c24ccaf89cbf67898cff53c95ae/pytorch_model.bin'. If you tried to load a PyTorch model from a TF 2.0 checkpoint, please set from_tf=True.
I also tried from_tf = True:
```
OSError Traceback (most recent call last)
in <cell line: 25>()
23 bnb_4bit_compute_dtype = dtype,
24 )
---> 25 model = AutoModelForCausalLM.from_pretrained(
26 model_name,
27 device_map = "sequential",
1 frames
/usr/local/lib/python3.10/dist-packages/transformers/modeling_utils.py in 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)
3351 )
3352 else:
-> 3353 raise EnvironmentError(
3354 f"{pretrained_model_name_or_path} does not appear to have a file named"
3355 f" {_add_variant(WEIGHTS_NAME, variant)}, {TF2_WEIGHTS_NAME}, {TF_WEIGHTS_NAME} or"
OSError: TinyLlama/TinyLlama-1.1B-intermediate-step-1431k-3T does not appear to have a file named pytorch_model.bin, tf_model.h5, model.ckpt or flax_model.msgpack.
Again great work!
can you share the code you used for loading the model :)
Interestingly I just noticed the stable release of transformers==4.36.2 works fine! It might be an issue only with 4.37.
Oh my code is literally the normal code for loading:
model = AutoModelForCausalLM.from_pretrained(
"TinyLlama/TinyLlama-1.1B-intermediate-step-1431k-3T",
torch_dtype = torch.bfloat16,
)
getting the same issue
@manishiitg I reuploaded it at https://huggingface.co./unsloth/tinyllama to work for the transformers dev branch!
This is caused by safe loading feature introduced by transformers==4.37, which add weights_only=True argument to the torch.load function. To fix this issue, you can simply load and re-save the model using the latest pytorch.
import torch
model = torch.load('pytorch_model.bin')
torch.save(model, 'pytorch_model.bin')
Can confirm the issue, can confirm that loading and saving fixes it.