AMD-AGI/Hyperloom

An agentic system that auto-optimizes LLM workloads on AMD GPUs.

Stars: 193
Forks: 54
Open beginner issues: 0
Indexed issues: 24
Avg merge: 1d 17h
Merged PRs (30d): 196

Dominant language: Python
License: No license data
Last GitHub push: Sep 18, 2026
Latest indexed: Sep 20, 2026
Contributing guide: Contributing guide
Code of conduct: Code of conduct
Beginner labels: No beginner labels indexed

Beginner friendly Everything

Open All

24 open issues indexed

optimization_journal.json records baseline_throughput = 0.0 for runs that measured a real baseline Open

Difficulty 3/5 1-2 days Newbie friendliness 72/100

AMD-AGI/Hyperloom#1554 · Opened Sep 17, 2026
integrate_patch is permanently denied under --critic-mock: the mock's approval never lands as a specialist patch verdict Open

Difficulty 4/5 3-5 days Newbie friendliness 58/100

AMD-AGI/Hyperloom#1553 · Opened Sep 17, 2026
critic/robustness agents (protocol=anthropic) appear to lack the idle timeout ClaudeBackend has — a stalled round wedges the coordinator Open

awaiting-author-response

AMD-AGI/Hyperloom#1552 · 1 comment · 1 assignee · Opened Sep 17, 2026
Scriptable / custom workloads get an empty EXPLORE grid — let an adapter declare its own tunable levers Open

Difficulty 3/5 1-2 days Newbie friendliness 35/100

AMD-AGI/Hyperloom#1544 · Opened Sep 17, 2026
emit_intent tool input arriving as a JSON string (litellm-style proxy) is silently dropped → autonomous loop stalls at 0% Open

Difficulty 2/5 1-3 hours Newbie friendliness 35/100

AMD-AGI/Hyperloom#1543 · Opened Sep 17, 2026
Proposal: sparse (IRT-calibrated) accuracy evaluations for the accuracy gate Open

AMD-AGI/Hyperloom#1539 · 1 assignee · Opened Sep 16, 2026
`isolated` vLLM: patches are applied to a copy that never serves Open

Difficulty 5/5 Over a week Newbie friendliness 35/100

AMD-AGI/Hyperloom#1521 · Opened Sep 16, 2026
TraceLens Bypass Path Cleanup Open

domain:inference type:feature

AMD-AGI/Hyperloom#1517 · 1 assignee · Opened Sep 15, 2026
KB warm replay does not compare the ROCm/AITER build a config was tuned on Open

domain:inference type:feature

AMD-AGI/Hyperloom#1507 · 1 comment · 1 assignee · Opened Sep 14, 2026
[Orchestrator] conversational roles stall at max_turns=12; kill+resume leaves phantom coordinator leases that deadlock IR-1 Open

Difficulty 5/5 Over a week Newbie friendliness 38/100

AMD-AGI/Hyperloom#1506 · 1 comment · Opened Sep 14, 2026
[Magpie/benchmark] Parent HIP_VISIBLE_DEVICES leaks into ROCR-pinned child env → server dies before port bind (proven reconcile patch available) Open

Difficulty 3/5 1-2 days Newbie friendliness 76/100

AMD-AGI/Hyperloom#1505 · 2 comments · Opened Sep 14, 2026
[Inference] enablement patches authored & proven, but integration lane never lands them → `enablement_stalled` with 0 baselines (gfx90a/MI250X) Open

Difficulty 4/5 3-5 days Newbie friendliness 48/100

AMD-AGI/Hyperloom#1504 · 2 comments · Opened Sep 14, 2026
install.sh: a user-supplied GEAK_ROOT working checkout is overwritten Open

type:bug

AMD-AGI/Hyperloom#1481 · 1 assignee · Opened Sep 10, 2026
Valid completed benchmarks are discarded when the wrapper script exits non-zero (magpie_nonzero_after_valid_measurement) Open

domain:inference type:bug

AMD-AGI/Hyperloom#1478 · 1 assignee · Opened Sep 10, 2026
--force-resume with a raised --target-gain closes immediately: target_reached_at survives the resume Open

domain:inference type:bug

AMD-AGI/Hyperloom#1469 · 1 assignee · Opened Sep 10, 2026
Track experimental Radeon 8065S (Halo/gfx1151) custom-workload bring-up Open

AMD-AGI/Hyperloom#1467 · 1 assignee · Opened Sep 10, 2026
Verify GEAK CI recovery: paused dispatch and no selected new contribution Open

AMD-AGI/Hyperloom#1440 · 1 assignee · Opened Sep 8, 2026
GEAK records acceptance after native current-best promotion is vetoed Open

AMD-AGI/Hyperloom#1436 · 1 assignee · Opened Sep 8, 2026
Retained GEAK overlays are omitted from concurrency sweep and exported launch script Open

AMD-AGI/Hyperloom#1435 · 1 assignee · Opened Sep 8, 2026
GEAK fallback can promote candidates rejected by native accuracy validation Open

AMD-AGI/Hyperloom#1433 · 1 assignee · Opened Sep 8, 2026
MiniMax MXFP4 SGLang drops SwiGLU-OAI semantics, blocking valid GEAK validation Open

AMD-AGI/Hyperloom#1432 · 1 assignee · Opened Sep 8, 2026
Preserve environment values across the GEAK handoff and return path Open

AMD-AGI/Hyperloom#1429 · 1 assignee · Opened Sep 7, 2026
RDNA support (gfx11/gfx12): interest in a contributed runner + profiling backend? Open

Difficulty 5/5 Over a week Newbie friendliness 32/100

AMD-AGI/Hyperloom#1196 · 8 comments · Opened Aug 17, 2026
Add RDNA4 GPU support (RX 9070 XT, RX 9070, RX 9060 XT, RX 9000 XT) Open

domain:inference domain:training type:feature

Difficulty 4/5 3-5 days Newbie friendliness 55/100

AMD-AGI/Hyperloom#1033 · 2 comments · Opened Jul 26, 2026

A short digest of beginner-friendly GitHub issues.