Pinned Loading

1. phantom phantom Public

   PHANTOM enables consumer laptops and single GPUs (6GB–8GB VRAM) to run 70B+ parameter models (LLaMA-3 70B, Qwen2 72B, Mixtral MoE) with long context windows by orchestrating a 3-tier memory hierar…

   Python 1

2. Reward-Hackability-Auditor--CLI---Claude-Skill- Reward-Hackability-Auditor--CLI---Claude-Skill- Public

   🐀 Fuzz your verifier before an RL agent does. Static + dynamic LLM security auditor to detect reward-hacking in RL post-training environments (OpenEnv, verifiers-spec, Gymnasium).

   Python 5

3. EvalGuard EvalGuard Public

   EvalGuard wraps existing evaluation harnesses (SWE-bench, Terminal-Bench, OpenEnv) without altering benchmark test suites or agent implementations. It audits sandbox boundaries, detects reward hac…

   Python 4

4. ONEFORALL ONEFORALL Public

   Let multiple developers and their AI agents work on the same repo without merge conflicts.

   JavaScript 3

5. RL-for-Universal-Compiler-Optimization RL-for-Universal-Compiler-Optimization Public

   Offline Reinforcement Learning for Universal Compiler Optimization across CPUs and GPUs using LLVM IR, MLIR, Graph Neural Networks, and hardware-aware reward modeling.

   Python 2

6. chessy chessy Public

   Host real-time AI chess tournaments in a live browser UI, audit custom models across 16 FIDE edge benchmarks, and referee agent matches in isolated sandboxes.

   Python 4