Pinned Loading

1. llm-finetuning llm-finetuning Public

   LoRA/PEFT fine-tunes for code and terminal agents - HumanEval 68.9%, Terminal-Bench 2.0 eval artifacts

   Python

2. llm-serving-benchmarks llm-serving-benchmarks Public

   Measurement-driven LLM inference optimization: raw 2xA100/H100 benchmarks, calibrated roofline model, and a self-optimizing serving loop (APAC HPC-AI prep)

   Python

3. matilda-llm matilda-llm Public

   Decoder-only LLMs trained from scratch in pure PyTorch - 114M to 152M, beats Pythia-160M, crash-safe training infrastructure

   Python

4. melbourne-cbd-crowd-backend melbourne-cbd-crowd-backend Public

   Real-time sensory-friendly navigation for Melbourne CBD. Flask API with LightGBM to ONNX crowd forecasts, calm-route planning, zero-PII auth

   Python

5. terminal-agent-sft terminal-agent-sft Public

   Terminal agent SFT pipelines: 366K-to-50K trajectory curation, TB2 decontamination, honest evaluation

   Python