Popular repositories Loading

1. unsloth-colab-finetuning unsloth-colab-finetuning Public

   Clean workspace for fine-tuning LLMs with Unsloth in Google Colab - GitHub-to-Colab sync, switchable custom JSONL datasets, ready-to-run training notebook

   Jupyter Notebook

2. GPT-light GPT-light Public

   GPT language model built and trained from scratch in plain PyTorch (97M params; RoPE, SwiGLU, RMSNorm, QK-norm, Muon optimizer) - pretrained on FineWeb-Edu with free Colab/Kaggle GPUs, with SFT, a …

   Jupyter Notebook

3. Say43 Say43 Public

   Profile README - github.com/Say43

4. AgentLight AgentLight Public

   Coding agent: Llama 3.2 3B fine-tuned with reasoning-SFT and GRPO on verifiable unit-test rewards, plus a ReAct execute-and-repair loop. Trained on free Kaggle GPUs. Built with Llama.

   Python

5. TurbofanRUL-NCMAPSS TurbofanRUL-NCMAPSS Public

   RUL prediction for turbofan engines on NASA N-CMAPSS (DS02): classical ML (NGBoost + conformal prediction) vs. deep learning (CNN deep ensemble) with calibrated uncertainty quantification.

   Jupyter Notebook