arinbalyan/smolvlm-chartqa-lora

SmolVLM ChartQA LoRA Adapter

Note: Training did not complete successfully.

This directory was set up as the model repo for a SmolVLM LoRA adapter fine-tuned on ChartQA. However, the Kaggle training notebook encountered an environment error (torchaudio version conflict with torch 2.5.1 on P100 GPU) and did not produce LoRA weights.

Status

• Training notebook: arinbalyan/smolvlm-fine-tune-on-chartqa
• Issue: OSError — libtorchaudio.so: undefined symbol: aoti_torch_memory_format_preserve_format caused by torchaudio 2.10.0+cu128 requiring torch==2.10.0 while the P100 environment pins torch==2.5.1+cu118.
• No LoRA weights were produced due to this environment conflict.

Next Steps

Rerun the notebook in an environment without the torchaudio conflict (e.g., disable torchaudio install or use compatible versions) to generate actual adapter_model.safetensors.