notSnix/Step-3.7-Flash-Q4_K_M-GGUF

Step 3.7 Flash Q4_K_M GGUF

This repo contains the full text-side GGUF quantization of stepfun-ai/Step-3.7-Flash.

For speculative decoding, use the companion MTP draft GGUFs here:

notSnix/Step-3.7-Flash-MTP-Draft-GGUF

The source model is Apache-2.0. The original model is multimodal, but this GGUF artifact was prepared and tested for text-side llama.cpp serving.

Files

File	Size	SHA256	Purpose
Step-3.7-Flash-Q4_K_M.gguf	111 GB	4de6519cf0131820d81137ebe6a0ab8dc225f1c463cc385038ab7de41ee7a36f	Full model
chat_template.jinja	5.6 KB	f428623fc81c940c35be3509fbffc086b4b4360d8800e46103e6f34d02891633	Chat template

Runtime

Current llama.cpp main supports Step MTP draft loading natively when used with the companion draft repo. This was smoke-tested with clean llama.cpp commit d545a2a993849fcf3b752d85ae256fc9d6a9de79.

git clone https://github.com/ggml-org/llama.cpp
cd llama.cpp
cmake -B build-cuda -DGGML_CUDA=ON
cmake --build build-cuda --config Release -j

Basic Command

llama-server \
  --model Step-3.7-Flash-Q4_K_M.gguf \
  --host 0.0.0.0 \
  --port 8000 \
  --ctx-size 262144 \
  --n-gpu-layers all \
  --split-mode layer \
  --parallel 1 \
  --reasoning on \
  --reasoning-format deepseek \
  --chat-template-file chat_template.jinja

With MTP Draft

Download an MTP draft GGUF from notSnix/Step-3.7-Flash-MTP-Draft-GGUF, then run:

llama-server \
  --model Step-3.7-Flash-Q4_K_M.gguf \
  --model-draft Step-3.7-Flash-MTP-Q8_0.gguf \
  --host 0.0.0.0 \
  --port 8000 \
  --ctx-size 262144 \
  --n-gpu-layers all \
  --split-mode layer \
  --parallel 1 \
  --reasoning on \
  --reasoning-format deepseek \
  --spec-type draft-mtp \
  --spec-draft-n-max 2 \
  --spec-draft-p-min 0.60 \
  --chat-template-file chat_template.jinja

Local Benchmark Snapshot

GPUs: RTX PRO 6000, 3x RTX 3090.

Recommended local MTP setting from the tested sweep: --spec-draft-n-max 2 --spec-draft-p-min 0.60 with the Q8_0 draft.

Run	Prompt tokens	Prefill	Decode	TTFT	Notes
Q4_K_M + Q8_0 MTP n_max=2 p_min=0.60	32,769	1823.47 tok/s	104.38 tok/s	18.054 s	87.1% draft accepted
Q4_K_M + BF16 MTP n_max=2 p_min=0.60	32,769	1835.66 tok/s	93.38 tok/s	17.904 s	79.3% draft accepted
Q4_K_M + BF16 MTP n_max=2 p_min=0.60	65,537	1626.84 tok/s	94.79 tok/s	40.391 s	81.2% draft accepted
Q4_K_M + MTP n_max=3	604	-	143.81 tok/s	0.415 s	172/181 draft accepted, 95.0%
Q4_K_M + MTP n_max=3	32,519	2097.79 tok/s	104.91 tok/s	15.62 s	60/73 draft accepted, 82.2%
Q4_K_M + MTP n_max=3	54,619	1909.23 tok/s	106.73 tok/s	28.82 s	60/70 draft accepted, 85.7%
Q4_K_S baseline	604	1738.12 tok/s	110.70 tok/s	0.352 s	no MTP
Q4_K_S baseline	54,619	2194.42 tok/s	89.15 tok/s	25.16 s	no MTP

Limited task checks:

Check	Q4_K_S baseline	Q4_K_M + MTP n_max=3
ARC Challenge chat, 10 samples	0.9	0.9
GSM8K strict/flexible, 10 samples	0.9 / 0.9	0.8 / 0.8
Code needle / NIAH reasoning-aware	12/12	12/12

Checksums

sha256sum -c SHA256SUMS

Notes

• The base model advertises 256k context; this GGUF release was loaded locally at 256k context.
• The MTP draft GGUFs are companion files for speculative decoding and are hosted separately to avoid confusing them with full-model quants.
• This is a community GGUF quantization/repackaging of the upstream Apache-2.0 model, not an official StepFun release.