Preview checkpoint (step 3,500) and model card

README.md

@@ -6,9 +6,11 @@ tags:
   - vllm
 ---
 
-# Kimi K3.1 DSpark
+> **Preview checkpoint.** We are working on a stronger version of this draft and will release it soon.
 
-DSpark draft model for Kimi K3.1.
+# Kimi K3.1 Pro DSpark (preview)
+
+DSpark draft model for Kimi K3.1 Pro.
 
 | Component | Value |
 | --- | --- |
@@ -22,7 +24,7 @@ DSpark draft model for Kimi K3.1.
 
 ## Evaluation settings
 
-- **Server:** `--speculative-config '{"method":"dspark","model":"Inferact/Kimi-K3.1-DSpark","num_speculative_tokens":7,"draft_sample_method":"probabilistic","rejection_sample_method":"block"}'`
+- **Server:** `--speculative-config '{"method":"dspark","model":"Inferact/Kimi-K3.1-Pro-DSpark-preview","num_speculative_tokens":7,"draft_sample_method":"probabilistic","rejection_sample_method":"block"}'`
 - **Client:** `vllm-bench` chat completions, temperature 1.0, top_p 0.95, full test set per
   benchmark.
 - **Metric:** acceptance length = 1 + accepted draft tokens / drafts (counts the bonus token).
@@ -31,20 +33,20 @@ DSpark draft model for Kimi K3.1.
 
 | Benchmark | Prompts | Max tokens | Acceptance length | Accept rate, positions 1-7 |
 | --- | ---: | ---: | ---: | --- |
-| GSM8K | 1,319 | 1,024 | **6.004** | 0.94 0.87 0.79 0.71 0.63 0.56 0.49 |
-| HumanEval | 164 | 1,024 | 5.424 | 0.91 0.81 0.71 0.61 0.53 0.46 0.39 |
-| MBPP | 974 | 1,024 | 4.873 | 0.89 0.76 0.64 0.53 0.43 0.35 0.28 |
-| MATH-500 | 500 | 2,048 | 4.149 | 0.82 0.65 0.51 0.40 0.32 0.25 0.20 |
-| SPEED-Bench low entropy (10K-token inputs) | 512 | 1,536 | 3.889 | 0.81 0.62 0.47 0.36 0.27 0.21 0.15 |
-| MT-Bench | 80 | 1,024 | 3.549 | 0.78 0.57 0.41 0.29 0.22 0.16 0.11 |
-| SWE-bench Pro | 642 | 2,048 | 3.379 | 0.77 0.56 0.39 0.27 0.19 0.12 0.08 |
-| AIME 2026 | 30 | 4,096 | 2.955 | 0.71 0.47 0.31 0.20 0.13 0.08 0.05 |
-| **Pooled** | | | **4.035** | 0.82 0.64 0.50 0.38 0.30 0.23 0.18 |
+| GSM8K | 1,319 | 1,024 | **4.597** | 0.86 0.71 0.59 0.48 0.39 0.32 0.26 |
+| HumanEval | 164 | 1,024 | 3.585 | 0.76 0.56 0.41 0.30 0.23 0.18 0.14 |
+| MBPP | 974 | 1,024 | 3.520 | 0.77 0.56 0.41 0.29 0.21 0.16 0.12 |
+| MATH-500 | 500 | 2,048 | 3.824 | 0.80 0.62 0.46 0.35 0.26 0.19 0.14 |
+| SPEED-Bench low entropy (10K-token inputs) | 512 | 1,536 | 2.930 | 0.70 0.46 0.30 0.19 0.13 0.09 0.06 |
+| MT-Bench | 80 | 1,024 | 2.812 | 0.69 0.44 0.27 0.17 0.11 0.07 0.05 |
+| SWE-bench Pro | 642 | 2,048 | 2.741 | 0.67 0.43 0.27 0.16 0.10 0.06 0.04 |
+| AIME 2026 | 30 | 4,096 | 2.973 | 0.71 0.49 0.32 0.20 0.13 0.08 0.05 |
+| **Pooled** | | | **3.154** | 0.72 0.50 0.34 0.23 0.16 0.11 0.08 |
 
 ## Long context
 
 | Benchmark | Prompts | Prompt tokens | Max tokens | Acceptance length | Accept rate, positions 1-7 |
 | --- | ---: | ---: | ---: | ---: | --- |
-| AA-LCR | 100 | 95K mean | 1,024 | **3.893** | 0.80 0.63 0.49 0.37 0.27 0.20 0.14 |
-| BEAM 128K | 400 | 97K-125K | 2,048 | 3.809 | 0.82 0.63 0.47 0.35 0.25 0.17 0.12 |
-| BEAM 1M | 700 | 826K-1.04M | 2,048 | 3.255 | 0.78 0.55 0.38 0.24 0.15 0.09 0.06 |
+| AA-LCR | 100 | 95K mean | 1,024 | **3.501** | 0.75 0.57 0.42 0.30 0.22 0.15 0.10 |
+| BEAM 128K | 400 | 97K-125K | 2,048 | 2.954 | 0.71 0.48 0.31 0.20 0.13 0.08 0.05 |
+| BEAM 1M | 700 | 826K-1.04M | 2,048 | pending | pending |

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:27ad5181c8ea09d125615796b3d7e279dfe4300207b895de84a107611533d027
+oid sha256:2fe8912aaff39d5c9603000fab4c4fe9da08aadf3e13b0c1e1fb0011949de4f7
 size 5373511978