samadpls/torchlight-threads
Viewer • Updated • 29.8k • 155 • 3
torchlight
PyTorch helpers trained on the answers that solved real forum threads: the dataset, and fine-tunes of Phi-4-mini and Qwen3-8B.
Note The dataset: PyTorch forum questions paired with the answers that solved them.
samadpls/torchlight-phi4-mini
Text Generation • 4B • Updated • 1.18k • 1Note The full model: Phi-4-mini-instruct fine-tuned with LoRA. Use with transformers or mlx-lm.
samadpls/torchlight-qwen3-8b
Text Generation • 8B • Updated • 182 • 1Note The newer, bigger model: Qwen3-8B fine-tuned with LoRA on a free Kaggle GPU. Fewer wrong answers than its base.