Natural Reasoning LMs
Collection
LMs fine tuned on natural reasoning by facebook
•
2 items
•
Updated
•
1
4bit-SmolLM2-natural-reasoning-10k
fine tuned on facebook/natural_reasoning for 10K steps on one RTX4060 using 4bit quantization.
Evaluated using LightEval
Framework versions
- Optimizer: AdaFactor
- Bitsandbytes
- PEFT 0.14.0
Results
| Dataset | Baseline | Ours |
|---|---|---|
| CommonsenseQA | 19.5 | 20.2 |
| PIQA | 3.1 | 12.4 |
| Winogrande | 54.6 | 54.8 |
| HellaSwag | 21.7 | 25.6 |
| MMLU | 20.2 | 19.3 |
Note: Scores Taken from here
- Downloads last month
- 102
Inference Providers
NEW
This model isn't deployed by any Inference Provider.
🙋
Ask for provider support
Model tree for chungimungi/4bit-SmolLM2-natural-reasoning-10k
Base model
HuggingFaceTB/SmolLM2-360M