Upload README.md with huggingface_hub

README.md

@@ -115,7 +115,9 @@ EAGLE3 trains a single-layer draft head that predicts the next token using hidde
 
 *The released model was fine-tuned for 6 additional epochs on 20K regenerated samples from the target model. The fine-tuned accuracy is expected to be equal or higher than these base values.*
 
-### Inference Benchmarks (B=1, temp=0, FP8, TP=4)
+### Inference Benchmarks (B=1, temp=0, TP=4)
+
+**With draft_tokens=8 (best B=1 config)**:
 
 | Dataset | Baseline (tok/s) | EAGLE3 (tok/s) | Speedup |
 |---------|-----------------|----------------|---------|
@@ -124,7 +126,18 @@ EAGLE3 trains a single-layer draft head that predicts the next token using hidde
 | SWEBench-Verified | 109.6 | 191.8 | **1.75x** |
 | Aider | 109.9 | 186.8 | **1.70x** |
 
-*Config: steps=3, topk=4, draft_tokens=8. All datasets at temp=0 on 8x H200 (TP=4).*
+*Config: steps=3, topk=4, draft_tokens=8. 8x H200 (TP=4).*
+
+**With draft_tokens=6 (standard config, verified 2026-04-12)**:
+
+| Dataset | Baseline (tok/s) | EAGLE3 (tok/s) | Speedup |
+|---------|-----------------|----------------|---------|
+| HumanEval | 109.6 | 158.0 | **1.44x** |
+| Terminal-Bench | 108.9 | 150.2 | **1.38x** |
+| MT-Bench | 109.0 | 143.6 | **1.32x** |
+| SWEBench-Verified | 109.1 | 116.5 | **1.07x** |
+
+*Config: steps=3, topk=4, draft_tokens=6. 4x H200 (TP=4). Server-side Prometheus metrics.*
 
 ## Model Architecture