Ling3-DSpark
面向 Ling3 的 DSpark 投机解码器。DSpark 在 DFlash 基础上扩展了目标模型辅助特征,并引入了一个置信度头,可动态选择草稿 token 的数量。该模型使用 SpecForge 训练,并通过 SGLang 提供服务。
模型规格
- 目标模型:Ling-3.0-flash
- 草稿参数量:1,363,707,905(1.36B)
- 草稿权重数据类型:BF16
- 隐藏层大小:2,560
- Transformer 层数:5 个全注意力层
- 注意力机制:MHA,32 个查询头,32 个键/值头
- 目标辅助特征层:1、11、23、29、35
- 置信度头:vanilla Markov 头,秩为 256
- DSpark 块大小:8 个草稿 token(验证宽度为 9,包含目标奖励 token)
- 最大位置嵌入:262,144
接受长度
接受长度是指每次投机验证步骤中平均被接受的 token 数量,包含目标奖励 token。
| 工作负载 | 接受长度 |
|---|---|
| GSM8K | 6.40 |
| MATH-500 | 6.29 |
| AIME 2025 | 5.56 |
| HumanEval | 6.57 |
| MBPP | 6.34 |
| LiveCodeBench | 5.33 |
| MT-Bench | 3.92 |
| Alpaca | 3.51 |
| Arena-Hard-v2 | 3.72 |
九个工作负载均值的宏观平均值为 5.29。
使用 SGLang 提供服务
使用支持 DSPARK 的 SGLang 版本。将模型路径和张量并行大小替换为适合您部署环境的值:
sglang serve \
--trust-remote-code \
--model-path <LING3_MODEL_PATH> \
--tp-size <TP_SIZE> \
--speculative-algorithm DSPARK \
--speculative-draft-model-path <LING3_DSPARK_MODEL_PATH> \
......
1B 参数
Ling3-DSpark
A DSpark speculator for Ling3. DSpark extends DFlash with target-model auxiliary features and a confidence head that dynamically chooses the number of draft tokens. The model was trained with SpecForge and is served with SGLang.
Model specifications
- Target model: Ling-3.0-flash
- Draft parameters: 1,363,707,905 (1.36B)
- Draft weight dtype: BF16
- Hidden size: 2,560
- Transformer layers: 5 full-attention layers
- Attention: MHA with 32 query heads and 32 key/value heads
- Target auxiliary feature layers: 1, 11, 23, 29, 35
- Confidence head: vanilla Markov head, rank 256
- DSpark block size: 8 draft tokens (verify width 9, including the target bonus token)
- Maximum position embeddings: 262,144
Acceptance length
Acceptance length is the mean number of tokens accepted per speculative verification step, including the target bonus token.
| Workload | Acceptance length |
|---|---|
| GSM8K | 6.40 |
| MATH-500 | 6.29 |
| AIME 2025 | 5.56 |
| HumanEval | 6.57 |
| MBPP | 6.34 |
| LiveCodeBench | 5.33 |
| MT-Bench | 3.92 |
| Alpaca | 3.51 |
| Arena-Hard-v2 | 3.72 |
The macro mean across the nine workload means is 5.29.
Serving with SGLang
Use an SGLang version with DSPARK support. Replace the model paths and tensor-parallel size with values appropriate for your deployment:
sglang serve \
--trust-remote-code \
--model-path <LING3_MODEL_PATH> \
--tp-size <TP_SIZE> \
--speculative-algorithm DSPARK \
--speculative-draft-model-path <LING3_DSPARK_MODEL_PATH> \
......
1B params