AMDResearch/ai4science-studio

How-to guides and AMD/ROCm optimization recipes for open AI-for-science models.

4Stars on the repository
53Mods indexed here, across every type
1mo agoLast push, which is what freshness is scored on
MITLicence, which decides whether bodies are shown

launcher

01

AMDResearch/ai4science-studio

Agent

Submits a 2-node ORBIT-2 training (AMD Instinct MI355X) with PyTorch profiling and Omnistat user-mode telemetry, waits for completion, and writes manifest.json for downstream subagents.

4 1mo ago A 0 tokens original MIT

omnistat_analyst

02

AMDResearch/ai4science-studio

Agent

Drive omnistat-inspect (PR #271) through the analyze-job phases on the user-mode VictoriaMetrics database, then map findings to the bottleneck taxonomy.

4 1mo ago A 0 tokens original MIT

omnistat_verifier

03

AMDResearch/ai4science-studio

Agent

Independently re-derive the top 2-3 claims from omnistat/claims.json by issuing raw PromQL via curl against the same VictoriaMetrics endpoint, at the finest sampling step. Optionally probe one cheap remedy on a 1-node interactive srun.

4 1mo ago A 0 tokens original MIT

AMDResearch/ai4science-studio

Agent

You are the GEMM-time bottleneck orchestrator for ORBIT-2 (Bayes-CAST EDM, 8M) on MI355X / ROCm 7.2.2 / PyTorch 2.10. You run unattended via the Claude Code CLI in tmux. Goal: produce a TraceLens + Omnistat analyst/verifier bottleneck analysis of where compute time goes — with a focus on the GEMMs — at 1 node and 2…

4 1mo ago A 0 tokens original MIT

synthesizer

05

AMDResearch/ai4science-studio

Agent

Merge tracelens/verifiedclaims.json and omnistat/verifiedclaims.json into a single ranked, deduplicated bottleneck report.

4 1mo ago A 0 tokens original MIT

tracelens_analyst

06

AMDResearch/ai4science-studio

Agent

Generate a TraceLens performance report for the rank-0 PyTorch trace and produce structured claims.json mapped to the bottleneck taxonomy.

4 1mo ago A 0 tokens original MIT

tracelens_verifier

07

AMDResearch/ai4science-studio

Agent

Independently re-derive the top 2-3 claims from tracelens/claims.json against the raw .pt.trace.json and (optionally) probe one cheap remedy on a 1-node interactive srun.

4 1mo ago A 0 tokens original MIT

fom_extractor

08

AMDResearch/ai4science-studio

Agent

Computes per-iteration FOMs for the ORBIT-2 loop and optionally TraceLens↔Omnistat correlation (reuse HydraGNN §4 algorithm when traces exist).

4 1mo ago A 0 tokens original MIT

lever_picker

09

AMDResearch/ai4science-studio

Agent

Same contract as materialscience/models/HydraGNN/recipes/perf-optimizer-loop/agents/leverpicker.md with these substitutions.

4 1mo ago A 0 tokens original MIT

orchestrator

10

AMDResearch/ai4science-studio

Agent

Drives the ORBIT-2 (Bayes-CAST EDM) optimizer loop on MI355X-class hardware. Dispatches launcher + leverpicker + fomextractor + perf-analysis analysts/verifiers/synth + storywriter. Owns accept/revert on throughputsamplespers, STATUS.txt, STOP, and broken-node exit 42 handling.

4 1mo ago A 0 tokens original MIT