Instructions to use Hikari07jp/DSpark-Gemma-4-31B-draft with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use Hikari07jp/DSpark-Gemma-4-31B-draft with Transformers:

# Use a pipeline as a high-level helper
from transformers import pipeline

pipe = pipeline("text-generation", model="Hikari07jp/DSpark-Gemma-4-31B-draft")

# Load model directly
from transformers import AutoTokenizer, AutoModel

tokenizer = AutoTokenizer.from_pretrained("Hikari07jp/DSpark-Gemma-4-31B-draft")
model = AutoModel.from_pretrained("Hikari07jp/DSpark-Gemma-4-31B-draft")

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use Hikari07jp/DSpark-Gemma-4-31B-draft with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "Hikari07jp/DSpark-Gemma-4-31B-draft"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "Hikari07jp/DSpark-Gemma-4-31B-draft",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Use Docker

docker model run hf.co/Hikari07jp/DSpark-Gemma-4-31B-draft

SGLang

How to use Hikari07jp/DSpark-Gemma-4-31B-draft with SGLang:

Install from pip and serve model

# Install SGLang from pip:
pip install sglang
# Start the SGLang server:
python3 -m sglang.launch_server \
    --model-path "Hikari07jp/DSpark-Gemma-4-31B-draft" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "Hikari07jp/DSpark-Gemma-4-31B-draft",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Use Docker images

docker run --gpus all \
    --shm-size 32g \
    -p 30000:30000 \
    -v ~/.cache/huggingface:/root/.cache/huggingface \
    --env "HF_TOKEN=<secret>" \
    --ipc=host \
    lmsysorg/sglang:latest \
    python3 -m sglang.launch_server \
        --model-path "Hikari07jp/DSpark-Gemma-4-31B-draft" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "Hikari07jp/DSpark-Gemma-4-31B-draft",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Docker Model Runner
How to use Hikari07jp/DSpark-Gemma-4-31B-draft with Docker Model Runner:
```
docker model run hf.co/Hikari07jp/DSpark-Gemma-4-31B-draft
```

DSpark-Gemma-4-31B-draft

A retrained + extended DFlash draft model for google/gemma-4-31B-it, built on top of z-lab/gemma-4-31B-it-DFlash (Apache-2.0).

Two variants are included:

variant	path	what it is	works with
DSpark (semi-AR markov)	repo root	retrained backbone + rank-256 semi-AR markov head, probabilistic drafting	vLLM 0.24.0 + our patch
backbone-only	`backbone-only/`	retrained backbone, drop-in z-lab replacement	stock vLLM + PR #41703 (same as z-lab head)

Measured results

Setup: target = nvidia/Gemma-4-31B-IT-NVFP4 (official NVFP4), RTX PRO 6000 Blackwell (96 GB), vLLM 0.24.0 + PR#41703, K=15, --kv-cache-dtype bfloat16, draft attention_backend=flash_attn. Throughput from vLLM /metrics counters (not SSE chunk counting), ABBA alternation, warm-up rounds discarded, paired per-round statistics. Prompts are train-disjoint (24 per domain).

Sampling T=1 (the regime chat/agent traffic actually runs), K=15, n=3 paired rounds

domain	DSpark tok/s	z-lab tok/s	diff
math	217.4	207.1	+5.0%
code	197.6	185.9	+6.3%
chat	91.8	88.5	+3.7%
toolcall	181.5	191.3	−5.1%
aggregate	151.6	144.7	+4.7% (CI excludes 0)

Greedy (T=0), K=15, n=5 paired rounds

Aggregate +0.7% — below our measured greedy noise floor (~1.5%), so we report this as a tie. Both heads reach ~4.06x over non-speculative decoding (171.7 vs 42.3 tok/s).

Which variant should I use?

Chat / general / code / math with sampling → root variant (semi-AR markov): +4.7% over z-lab.
Tool-calling-heavy agents → backbone-only/: the markov bias hurts structured JSON (−5.1% vs z-lab on toolcall) while backbone-only is +6.6% over z-lab on toolcall.
You don't want to patch vLLM beyond PR#41703 → backbone-only/ (drop-in).

Quality

Speculative decoding is lossless by construction; we additionally verified end-to-end: gsm8k (n=150) and structured tool-call JSON (n=24) accuracy identical to non-speculative baseline (McNemar exact p=1.0, Δ=0.00pp), and NVFP4 target quality matches BF16 on the same battery.

Usage

backbone-only (stock DFlash path)

uv pip install -U --torch-backend=auto \
  "vllm @ git+https://github.com/vllm-project/vllm.git@refs/pull/41703/head"

vllm serve nvidia/Gemma-4-31B-IT-NVFP4 \
  --kv-cache-dtype bfloat16 \
  --speculative-config '{"method":"dflash","model":"Hikari07jp/DSpark-Gemma-4-31B-draft/backbone-only","num_speculative_tokens":15,"attention_backend":"flash_attn"}'

DSpark semi-AR (root variant)

Requires the DSpark patch on top of vLLM 0.24.0 — see hikarioyama/dspark-gemma4-31b for the patch and instructions. Then add "draft_sample_method":"probabilistic" to the speculative config and serve the repo root as the draft model.

Note: --kv-cache-dtype bfloat16 is required (FP8 KV is incompatible with the flash_attn draft path for this hybrid-SWA model), and FlashInfer rejects gemma-4's hybrid SWA — the config above is the known-good combination.

Training summary

Warm-started from z-lab/gemma-4-31B-it-DFlash, jointly retrained backbone + markov head
Loss: 0.9 L1 (hidden matching) + 0.1 CE, γ=6 position discounting, block 16, ctx 1024
15.5k on-policy samples generated by gemma-4-31B-it (chat-heavy mix: 50/35/15 chat/code/math in the general slice + tool-call + agent-trajectory sources)
6000 steps DDP; best checkpoint selected at step 2400 by held-out paired accept-length
Known corpus bias (honest): code slice is Python-heavy (51%) and implement-heavy (68%)

Acknowledgements

z-lab DFlash (paper) — the base drafter and the DFlash method
vLLM PR #41703 — gemma-4 DFlash batched-verification fixes this build depends on
Gemma is subject to the Gemma Terms of Use

Downloads last month: 451

Safetensors

Model size

2B params

Tensor type

BF16

Model tree for Hikari07jp/DSpark-Gemma-4-31B-draft

Base model

z-lab/gemma-4-31B-it-DFlash

Finetuned

(1)

this model

Paper for Hikari07jp/DSpark-Gemma-4-31B-draft

DFlash: Block Diffusion for Flash Speculative Decoding

Paper • 2602.06036 • Published Feb 5 • 89