lab-benchmarks / README.md
artificial-citizen's picture
Add vram_gb to all rows (artifact weight-file size) — unblocks /lab quality-vs-VRAM chart (protoLab#14, protoContent#394)
de1ce89 verified
|
Raw
History Blame Contribute Delete
1.05 kB
metadata
license: cc-by-4.0
tags:
  - benchmarks
  - quantization
  - nvfp4
  - mtp
  - llm-serving

protoLabs lab-benchmarks

Every number on a protoLabsAI model card traces to a row here. Release-gate results (quant vs bf16 baseline, paired task sets, outliers re-trialed x3 both sides), speed-test-v2 regime matrices (InferenceMAX-style: seeded random dataset, client-side TTFT/TPOT p50/p99, goodput), decode-at-depth ladders, and coherence-probe verdicts.

Methodology: single-stream-only numbers are never published without load numbers; spec-decode accept% is reported per workload (random-data benches understate it ~2.5x); LLM-judged suites are gated behind deterministic ones. Harness: protoLabsAI/protoLab evals.

vram_gb on each row is the artifact's published weight-file size in GB (the download you load) — the traceable X-axis for the quality-vs-VRAM chart. GGUF rows use the specific quant variant's file (e.g. the NVFP4 gguf, not a Q8_0 in the same repo), not the repo total.

CC-BY-4.0 — cite, steal, argue. Charts: protolabs.studio/lab