# Nokast: spark-s1-4b-v6-nvfp4

> NVFP4 (MLP-only) build of spark-s1-4b-v6 for vLLM on Blackwell-class GPUs: 1.4x faster than bf16 with no measurable JevBench public-item cost.

- Page: https://systemonemodels.ai/open-spark-jev/spark-s1-4b-v6-nvfp4
- API: https://api.systemonemodels.ai/v1/models/open-spark-jev/spark-s1-4b-v6-nvfp4
- Download: `pip install systemonemodels && systemone pull open-spark-jev/spark-s1-4b-v6-nvfp4`

## Facts

| | |
|---|---|
| Maker | Nokast (https://systemonemodels.ai/open-spark-jev) |
| Decides | choice, score, noul |
| Architecture | spark-s1 |
| Base model | open-spark-jev/spark-s1-4b-v6 |
| Licence | apache-2.0 |
| Availability | Open weights |
| Latest version | 0.1.0 |

## Reported evaluation

Suite: JevBench v1.2 public items only (231 of 534); accuracy on the public items, NOT the official JevBench Score. Numbers are the publisher's own.

- Decision accuracy: 81.0%

- Median latency: 53.3 ms

## Model card

## Weights

The files are hosted on Hugging Face and pinned to the revisions below (see `artifacts` in the manifest for hashes). Use `calibration.json` (per-question-type temperatures) with the option-letter logits.

* [abhishek085/spark-s1-4b-v6-nvfp4](https://huggingface.co/abhishek085/spark-s1-4b-v6-nvfp4/tree/13fae85becaeb909ac34ab7fa8255433624f0530)

Calibration temperatures are inherited from the bf16 model, so `calibration.json` is taken from `abhishek085/spark-s1-4b-v6`.

This is a decision model, not a chat model: see the usage notes in the card below.

> **About the JevBench numbers in this card.** Figures labelled "JevBench Intelligence" (83.1, 83.2, 82.2) are accuracy on the 231 public items only, mapped to a 0-100 scale. They are not the official JevBench Score. The only official result for these models is for `spark-s1-4b-v6` on JevBench v1.6.1: Capability 54.0, Composite 44.9, Intelligence 43.2 ([Benchmark Heaven](https://benchmarkheaven.com/api/decision-results/jevbench/v1.6.1/json)).


# spark-s1-4b-v6-nvfp4

NVFP4 (MLP-only) post-training quantization of [`abhishek085/spark-s1-4b-v6`](https://huggingface.co/abhishek085/spark-s1-4b-v6), part of
[Open Spark Jev](https://github.com/abhishek085/open-spark-jev). Only the MLP `gate_proj`/`up_proj`/`down_proj` weights are quantized to NVFP4;
every attention-related module (both the 8 full-attention layers and the 24 linear-attention layers — their conv1d, gating and projection
weights) plus `lm_head` stay in bf16, since the option-letter logit readout needs full precision there.

This is a very early release; expect it to change quickly.

## Speed vs accuracy

Measured on one NVIDIA DGX Spark (GB10), batch size 1, served via vLLM.

| | p50 | decisions/s | JevBench Intelligence |
|---|---:|---:|---:|
| `spark-s1-4b-v6` (bf16) | 74.9 ms | 13.3 | 83.1 |
| **`spark-s1-4b-v6-nvfp4`** | **53.3 ms** | **18.6** | **83.2** |

1.40x faster with **no measurable accuracy cost** — the small standard-tier dip (1.000 -> 0.972) is more than offset by a hard-tier gain
(0.595 -> 0.622, within the noise floor for 111 items). This is a much cleaner quantization result than `spark-s1-4b-v5-nvfp4`, whose same
recipe cost about 1.9 Intelligence points on the uniform Qwen3-4B backbone.

| Model | easy (48) | standard (72) | hard (111) |
|---|---:|---:|---:|
| `spark-s1-4b-v6-nvfp4` | 1.000 | 0.972 | 0.622 |
| `spark-s1-4b-v6` (bf16) | 1.000 | 1.000 | 0.595 |

[JevBench](https://github.com/fstandhartinger/jevbench) v1.2, public items only (231 of 534). This is the only accuracy check run on this
variant so far — not own-splits/external tested independently (inherits `spark-s1-4b-v6`'s own numbers there as a reasonable prior, given the
near-zero JevBench cost, but hasn't been directly measured).

The smaller speedup than v5-4b-nvfp4's 1.66x is expected: only the MLP is quantized, and this hybrid backbone carries relatively more of its
total compute in the unquantized attention/linear-attention paths. Even so, `spark-s1-4b-v6-nvfp4` (53.3ms) is faster than `spark-s1-4b-v5-nvfp4`'s
own base model `spark-s1-4b-v5` bf16 (61.6ms) while being far more accurate.

## How to serve

```bash
docker run --rm --gpus all -p 8355:8355 vllm/vllm-openai:nightly-aarch64 \
  --model abhishek085/spark-s1-4b-v6-nvfp4 --served-model-name spark-s1-4b-v6-nvfp4 \
  --host 0.0.0.0 --port 8355 --max-model-len 8192 --trust-remote-code
```

Then point the repo's gateway at it in OpenAI-backend mode:

```bash
python -m open_spark_jev.serve.gateway --backend openai --upstream http://127.0.0.1:8355/v1 --model spark-s1-4b-v6-nvfp4 --port 8472
```

`--max-model-len` should be at least ~4-8k tokens: some Jev-style decision states (long documents, tool traces) run well past 2k tokens.

## Reproduce this quantization

```bash
pip install nvidia-modelopt   # a current release (0.46+); the TensorRT-LLM container's bundled version predates Qwen3.5 support
python scripts/quant/ptq_nvfp4.py --model checkpoints/v6-4b --tokenizer models/Qwen3.5-4B \
  --calib runs/quant/calib_prompts_v5.jsonl --out engines/spark-s1-4b-v6-MLP_ONLY_CFG --cfg NVFP4_MLP_ONLY_CFG
```

Calibrated on the same 980 real decision prompts used for `spark-s1-4b-v5-nvfp4` (v6 trained on identical data, so these are directly reusable).

## Limitations

Inherits `spark-s1-4b-v6`'s limitations (Jev-style scope only, no RLCD, option-order sensitivity, English only). Additionally:

* Calibration temperatures in `calibration.json` are inherited from the bf16 checkpoint, not refitted on quantized outputs.
* Only JevBench-checked so far — own-splits and the 5-set external battery have not been independently re-measured on the quantized weights.
* Needs a GB10/B200-class (Blackwell) GPU for the NVFP4 tensor core path.

## License

Apache-2.0. Base model `Qwen/Qwen3.5-4B` (Apache-2.0).

---

From System One Models — https://systemonemodels.ai/ · every System One model: https://systemonemodels.ai/system-one-models