Back to search
Top Contributors
11
commits
8
commits
Agents, and RL environment, for optimizing GPU kernels on AMD ROCm using LLM agents. Benchmarks LLM serving workloads end-to-end, profiles bottleneck kernels, optimizes them via Claude Code or Codex, and scores on compilation, correctness, and speedup.
Stars
59
Forks
7
Watchers
59
Open Issues
2
Overall repository health assessment
No package.json found
This might not be a Node.js project
11
commits
8
commits
Merge pull request #6 from AMD-AGI/feat/gated-reward-pipeline
5725dadView on GitHubfix(graders): align detect_runtime_hacking() with Keystone checks
5f00380View on GitHubfeat(graders): wire gated reward into kernel grading pipeline
5ff8324View on GitHubfeat(graders): add compute_reward_gated() wrapper to score.py
00074e1View on GitHubfeat(graders): add difficulty weight schedule for RL curriculum
0a35a96View on GitHubfeat(graders): add Triton/HIP static analysis for reward hacking detection
14d98e2View on GitHubMerge pull request #4 from AMD-AGI/feat/knowledge-base-pipeline-hardening
3f9b262View on GitHubMerge pull request #3 from AMD-AGI/feat/correctness-modes-and-standalone-ops
ac03d04View on GitHubAdd testcase generator fixes, reinjection tests, and optional CK build for Accordo validation
32a38d7View on GitHubHarden hot-patch locking and logging; expand knowledge base and add tampering tests
88bab18View on GitHub