Results of run r2
Browse files
README.md
CHANGED
|
@@ -87,29 +87,32 @@ before choosing it:
|
|
| 87 |
|
| 88 |
## Results
|
| 89 |
|
| 90 |
-
Official kit `eval` (vLLM 0.31.0, all 160 dev prompts, K = 3), every row measured
|
| 91 |
-
run on the same **NVIDIA RTX PRO 6000 Blackwell**:
|
| 92 |
|
| 93 |
| draft | τ | chat / code / math / long-form | TPOT | memory | score, same GPU | score, published baseline |
|
| 94 |
|---|---|---|---|---|---|---|
|
| 95 |
-
| baseline (the target's built-in MTP head) | 2.
|
| 96 |
-
| this draft,
|
| 97 |
-
| **this draft, FP8 (submitted variant)** | **2.868** | 3.08 / 3.78 / 3.68 / 2.45 | 15.27 ms | **0.607 GB** | **114.4** | **110.7** |
|
| 98 |
|
| 99 |
* *Score, same GPU* uses the baseline as measured above; *score, published baseline* is the
|
| 100 |
kit's own number, whose speed term compares against the organisers' TPOT from different
|
| 101 |
hardware (that is why the baseline itself gets 96.2 there).
|
| 102 |
-
* The
|
| 103 |
-
|
| 104 |
-
+0.
|
| 105 |
-
* The offline simulator predicted the dev acceptance within 0.
|
| 106 |
-
2.
|
| 107 |
-
*
|
| 108 |
-
validation τ was still rising
|
| 109 |
-
|
| 110 |
-
|
| 111 |
-
|
| 112 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
| 113 |
|
| 114 |
## Layout
|
| 115 |
|
|
|
|
| 87 |
|
| 88 |
## Results
|
| 89 |
|
| 90 |
+
Official kit `eval` (vLLM 0.31.0, all 160 dev prompts, K = 3), run **r2**, every row measured
|
| 91 |
+
in the same run on the same **NVIDIA RTX PRO 6000 Blackwell**:
|
| 92 |
|
| 93 |
| draft | τ | chat / code / math / long-form | TPOT | memory | score, same GPU | score, published baseline |
|
| 94 |
|---|---|---|---|---|---|---|
|
| 95 |
+
| baseline (the target's built-in MTP head) | 2.805 | 3.07 / 3.77 / 3.63 / 2.36 | 15.39 ms | 0.917 GB | 100.0 | 96.2 |
|
| 96 |
+
| **this draft, FP8 (submitted variant)** | **2.900** | 3.08 / 3.77 / 3.70 / 2.50 | 15.24 ms | **0.607 GB** | **115.4** | **111.6** |
|
|
|
|
| 97 |
|
| 98 |
* *Score, same GPU* uses the baseline as measured above; *score, published baseline* is the
|
| 99 |
kit's own number, whose speed term compares against the organisers' TPOT from different
|
| 100 |
hardware (that is why the baseline itself gets 96.2 there).
|
| 101 |
+
* The same-GPU score splits into acceptance 1.053 (×0.5), memory 1.5 (×0.25, the cap) and
|
| 102 |
+
speed 1.010 (×0.25). Fine-tuning added +0.095 to τ, mostly in long-form writing (+0.14) and
|
| 103 |
+
math (+0.07); chat and code were already close to what one head layer can reach.
|
| 104 |
+
* The offline simulator predicted the dev acceptance within 0.02 of the engine (2.884 vs
|
| 105 |
+
2.900 for the draft, 2.793 vs 2.805 for the baseline).
|
| 106 |
+
* r2 trained on 4.06 M generated tokens (9,634 answers) for 1,190 steps (2 epochs, 27 min);
|
| 107 |
+
validation τ was still rising at the last step. The whole pipeline, including two full
|
| 108 |
+
evaluations, took about 3 hours of GPU time.
|
| 109 |
+
|
| 110 |
+
Earlier run **r1** (2.02 M tokens, 584 steps) reached τ 2.868 and a same-GPU score of 114.4
|
| 111 |
+
(bf16 variant: τ 2.885, score 102.3). Twice the data gave +0.03 τ.
|
| 112 |
+
|
| 113 |
+
Raw kit outputs, per-stage logs, machine statistics and training summaries are in the
|
| 114 |
+
dataset repo under `runs/r1/` and `runs/r2/`; the r2 draft folders (FP8 and bf16, each with
|
| 115 |
+
`model.safetensors`) under `submissions/r2/`.
|
| 116 |
|
| 117 |
## Layout
|
| 118 |
|