---

base_model: inclusionAI/Ling-3.0-flash-dspark

tags:

• transformers
• safetensors
• qwen3
• speculative-decoding
• dspark
• dflash
• specforge
• sglang
• text-generation
• license:other
• text-generation-inference
• endpoints_compatible
• region:us

---

Ling-3.0-flash DSpark GGUF

GGUF quantizations of inclusionAI DSpark draft model for Ling-3.0-flash.

Use with BeeLlama.cpp, a llama.cpp fork with advanced quantization features.

---

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

Launch recipes for this draft on every supported hardware/quantization cell — including the required --linear-replayssm-cache-len sizing — with measured speed and accuracy, are in the SGLang Ling-3.0-flash cookbook.

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> \
  ......

Serving with llama.cpp

Use a llama.cpp build with DSpark support. Replace the model paths, quantization type, and GPU layer counts with values appropriate for your deployment. First convert

and quantize the target model:

python convert_hf_to_gguf.py path/to/Ling-3.0-flash \
    --outfile path/to/Ling-3.0-flash-bf16.gguf \
    --outtype bf16 --model-name Ling-3.0-flash

llama-quantize \
    path/to/Ling-3.0-flash-bf16.gguf \
    path/to/Ling-3.0-flash-Q4_K_M.gguf \
    Q4_K_M

Then generate the DSpark draft GGUF:

python convert_hf_to_gguf.py \
    path/to/Ling-3.0-flash-dspark \
    --target-model-dir path/to/Ling-3.0-flash \
    --outtype bf16 \
    --outfile path/to/Ling-3.0-flash-DSpark.gguf

Finally, launch the server with the DSpark draft as the speculative model:

llama-server \
    --model path/to/Ling-3.0-flash-Q4_K_M.gguf \
    --spec-draft-model path/to/Ling-3.0-flash-DSpark.gguf \
    --spec-type draft-dspark --spec-draft-n-max 8 \
    -ngl all -ngld all -fa on