Upload folder using huggingface_hub

README.md

@@ -337,11 +337,9 @@ Per-kernel Nsight Compute (ncu) profiles across 4 GPUs (A100, H100, RTX 3090, RT
 
 Curated H100 / Llama-3.1-8B / vLLM validation table for the distributional synthetic replay generator. Paired synthetic and real trace replay runs; supplementary rows preserve no-replacement and high-concurrency debug runs. Raw JSON artifacts referenced through R2 URI columns. Per-run successful/failed request counts retained.
 
-### Concurrency filtering
+### Quality filtering
 
-trace_replay: concurrency > 100 removed. synthetic_distributional: concurrency > 10 removed.
-
-Configurations where fewer than 75% of requests completed successfully are excluded. Summary metrics are computed from successful requests only.
+Concurrency levels: trace_replay {1, 5, 10, 20, 40, 80}, synthetic_distributional {1, 5, 10}. Configurations where fewer than 75% of requests completed successfully are excluded. Summary metrics are computed from successful requests only.
 
 | Config | Rows |
 |--------|------|