RuntimeError: PytorchStreamReader failed reading file data/0: invalid header or archive is corrupted

[vip-china]

Please check that this issue hasn't been reported before.

• I searched previous Bug Reports didn't find any similar reports.

Expected Behavior

Generate the correct Lora after training is completed

Current behaviour

Using the following command to merge models, there is an error message：

python3 -m axolotl.cli.merge_lora sft_34b.yml --lora_model_dir="/workspace/axolotl/output/Yi-34B/ljf-yi-34b-lora" --output_dir=/data1/ljf2/data-check-test

Steps to reproduce

The meaning of this parameter is not effective
save_safetensors: true
Actually generated
adapter_model.bin

Config yaml

No response

Possible solution

No response

Which Operating Systems are you using?

• Linux
• macOS
• Windows

Python Version

3.10

axolotl branch-commit

main

Acknowledgements

• My issue title is concise, descriptive, and in title casing.
• I have searched the existing issues to make sure this bug has not been reported yet.
• I am using the latest version of axolotl.
• I have provided enough information for the maintainers to reproduce and diagnose the issue.

[winglian]

is the issue that the merge doesn't work? or that specifying save_safetensors produces a Pytorch_model.bin? or both?

[vip-china]

specifying save_safetensors produces a Pytorch_model.bin

[NanoCode012]

Using the following command to merge models, there is an error message：

Hey, seems like the PeftModel loading failed. Can you check the files in lora_model_dir are valid (aren't a few KBs)?

Did you run out of space during training?

Since the training had a few checkpoints, could you try pointing the model_dir to one of those and see what happens?

The meaning of this parameter is not effective
save_safetensors: true

I'm a bit confused at this as the you said the merge failed.

[vip-china]

Now I am continuing SFT training from checkpoint and reporting this error again

I have configured this parameter：
use_reentrant: true
resume_from_checkpoint: /workspace/axolotl-main/checkpoint-5865

[winglian]

resuming from a "peft checkpoint" is not the same as resuming from a regular checkpoint. You'll want to set lora_model_dir to point to the checkpoint directory iirc. @NanoCode012 does that sound right?