boltmonkey_shortreasoning‑8b
Merged QLoRA adapter plus base weights.
Fine‑tuned for short‑form chain‑of‑thought reasoning.
- Base model: SuperNeuralDreadDevil-8b
- Dataset: Cosmopedia‑Instruct 60k (ShareGPT style)
- Context length: 1096 tokens
- Training: 4 epochs, LoRA r = 32, α = 16, dropout 0.05, fp16, 4‑bit quant
See train_args.json for the full Axolotl config.
- Downloads last month
- 17