Skip to content

docs: experiment design justification - #35

Merged
sriumcp merged 3 commits into
inference-sim:mainfrom
susiejojo:docs/experiment-design-justification
Mar 16, 2026
Merged

sriumcp merged 3 commits into
inference-sim:mainfrom
susiejojo:docs/experiment-design-justification

Conversation

@susiejojo

Copy link
Copy Markdown
Contributor

Summary

  • Adds a concise justification document explaining why the ~50 experiments in Discussion #598 were chosen
  • Covers model selection (8 architectures, dense + MoE + FP8), hardware tiers (H100/A100/L40S), ServeGen workload profiles, config knob sweeps, and pruning rationale
  • Includes Vidur model overlap for head-to-head comparison and phase ordering for early-stopping guarantees

Context

Test plan

🤖 Generated with Claude Code

susiejojo and others added 3 commits March 16, 2026 16:35
…k matrix

Explains why these specific models, GPUs, workloads, and config knobs
were chosen for the BLIS validation experiments. Covers pruning rationale,
figure-driven mapping, phase ordering, Vidur overlap, and ServeGen workload
profiles. Written for executive/reviewer audience.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
@sriumcp
sriumcp merged commit 0f19016 into inference-sim:main Mar 16, 2026
1 check passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants