Pinned Loading

1. adaptive-test-time-compute adaptive-test-time-compute Public

   Adaptive Test-Time Compute Controller — per-instance compute allocation for LLM reasoning. 47% compute reduction at equivalent accuracy on GSM8K (n=300).

   Python 1

2. trellis-webgpu trellis-webgpu Public

   First trellis-coded LLM quantization decoder outside CUDA — an 8B model running its full forward pass in a browser on WebGPU. SOTA 3-bit quality (1.15x fp16).

   Python 2

3. agent-quality-inspect agent-quality-inspect Public

   Toolkit for user-aware evaluation of agentic AI systems: subgoal metrics (AUC, PPT, pass@k/pass^k) and automated LLM error analysis

   Python 1

4. gptq-from-scratch gptq-from-scratch Public

   From-scratch PyTorch implementation of GPTQ post-training quantization (2/3/4-bit) with multi-architecture support

   Python 1

5. onnx-graph-diff onnx-graph-diff Public

   CLI tool to locate N:M subgraph differences between ONNX models, with pluggable LLM-assisted matching and an independent equivalence checker

   Python 1

6. triton-kernels triton-kernels Public

   High-performance GPU kernels for LLM inference in OpenAI Triton: RMSNorm, SwiGLU, W8A16/W4A16 quantized GEMMs (symmetric + asymmetric), fused MoE dispatch

   Python 1