LoRA adapter for Smoltaur-0.1B, fine-tuned on a stratified 50% subset of Psych-101 as part of the LoRA-rank sweep and dataset-size ablation for Small Foundation Models of Human Cognition and Behaviour.
| field | value |
|---|---|
| base model | unsloth/SmolLM2-135M |
| LoRA rank | 16 (alpha = rank, rsLoRA) |
| data fraction | 50% of Psych-101 |
| training | 1 epoch, completion-only loss, seed 3407 |
Load with PEFT on top of unsloth/SmolLM2-135M, or evaluate with the project's
eval_model.py --backend unsloth.
adapter_model.safetensors 20 MBcheckpoint-500/adapter_model.safetensors 20 MBcheckpoint-940/adapter_model.safetensors 20 MBcheckpoint-500/optimizer.pt 11 MBcheckpoint-940/optimizer.pt 11 MBcheckpoint-500/rng_state.pth 14.6 kBcheckpoint-940/rng_state.pth 14.6 kBcheckpoint-500/training_args.bin 5.8 kBcheckpoint-940/training_args.bin 5.8 kBtraining_args.bin 5.8 kBcheckpoint-500/scheduler.pt 1.5 kBcheckpoint-940/scheduler.pt 1.5 kB有文件?提交种子。
提交种子unsloth/SmolLM2-135M