---
title: NVIDIA-Nemotron-3.5-Lightning-30B-A3B-DSpark-bf16
canonical_url: "https://www.modelscope.cn/models/mlx-community/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-DSpark-bf16"
md_url: "https://www.modelscope.cn/models/mlx-community/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-DSpark-bf16.md"
repository: mlx-community/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-DSpark-bf16
last_updated: 2026-08-11
license: other
pipeline_tag: text-generation
tasks:
  - text-generation
model_type:
  - qwen3
architectures:
  - Qwen3DSparkModel
base_model:
  - nvidia/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-NVFP4-DSpark
  - nvidia/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF16
base_model_relation: finetune
parameters: 967.6M
tensor_type:
  - BF16
library_name:
  - mlx
  - safetensors
  - pytorch
frameworks:
  - pytorch
downloads: 11
stars: 0
tags:
  - mlx
  - mlx-dspark
  - dspark
  - speculative-decoding
  - nemotron
---

# NVIDIA-Nemotron-3.5-Lightning-30B-A3B-DSpark-bf16

> NVIDIA-Nemotron-3.5-Lightning-30B-A3B-DSpark-bf16 - mlx-community 在 ModelScope 开源的模型。NVIDIA-Nemotron-3.5-Lightning-30B-A3B-DSpark (bf16, MLX)

mlx-community/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-DSpark-bf16 是 ModelScope 魔搭社区上的 967.6M 参数text-generation模型，采用 other 许可，基于 nvidia/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-NVFP4-DSpark、nvidia/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF16 构建。

- **Repository**: mlx-community/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-DSpark-bf16
- **License**: other
- **Tasks**: text-generation
- **Parameters**: 967.6M
- **Base model**: nvidia/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-NVFP4-DSpark, nvidia/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF16
- **Tags**: mlx, mlx-dspark, dspark, speculative-decoding, nemotron
- **Downloads**: 11
- **Stars**: 0
- **Last updated**: 2026-08-11

Source: https://www.modelscope.cn/models/mlx-community/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-DSpark-bf16

---

# NVIDIA-Nemotron-3.5-Lightning-30B-A3B-DSpark (bf16, MLX)

The **DSpark speculative-decoding drafter** for
[`nvidia/NVIDIA-Nemotron-3.5-Lightning-30B-A3B`](https://huggingface.co/nvidia/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF16),
dequantized to **bf16** for use on Apple Silicon with
[**mlx-dspark**](https://github.com/ARahim3/mlx-dspark).

This is a **format conversion, not a retrain.** NVIDIA ships the drafter as
[`nvidia/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-NVFP4-DSpark`](https://huggingface.co/nvidia/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-NVFP4-DSpark)
in NVIDIA Model Optimizer **NVFP4** (`W4A16_NVFP4`), which stock MLX / mlx-lm cannot read. The
FP4 (E2M1) weights + FP8 (E4M3) block scales were hand-decoded back to bf16 (attention /
`embed_tokens` / `markov_w1` were already bf16 in the source). Weights are otherwise unchanged;
mlx-dspark re-quantizes the drafter to 4-bit at load, which does not affect acceptance.

## Use with mlx-dspark

Pair it with the 4-bit MLX target
[`mlx-community/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-4bit`](https://huggingface.co/mlx-community/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-4bit):

```bash
pip install mlx-dspark
mlx-dspark generate \
  --model mlx-community/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-4bit \
  --drafter mlx-community/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-DSpark-bf16 \
  --max-draft 4 --prompt "Write a Python function for the n-th Fibonacci number."
```

The output is **lossless** — the target verifies every token, so speculation only changes speed,
never the text (it matches greedy target decoding up to floating-point ties).

## Architecture

A plain **Qwen3 GQA** drafter backbone (6 layers, hidden 2688, 32 heads / 2 KV, head_dim 128,
vocab 131072) with the DSpark **rank-512 Markov fixup head** and DFlash-lineage traits: **causal
sliding-window (1024) block attention** with a **per-head attention-sink bias**, `block_size 8`,
`sample_from_anchor=false` (predictions start at block slot 1). It ships its own `embed_tokens`
and **reuses the target's `lm_head`** (`has_lm_head=false`). 967M params total (615M
non-embedding). Fuses 6 target hidden-state taps (layers 1, 5, 19, 29, 41, 51).

The target is a hybrid **Mamba-2 + latent-MoE + attention** model (`model_type: nemotron_h`).
mlx-dspark adds the nemotron_h hidden-state tap and an exact Mamba-2 recurrent-state rollback so
the lossless verify loop works on it.

## Measured (M4 Pro, mlx 0.32, greedy, warm)

Baseline (4-bit target) ≈ **91 tok/s**. DSpark at `--max-draft 4` (lookup off):

| content | speedup | accept length |
|---|---|---|
| code | **1.27×** (→ ~117 tok/s) | 4.55 |
| math | **1.24×** | 4.41 |
| chat | 1.06× | 3.74 |

Throughput peaks at cap 4 and regresses past it (the MoE verify-width cost climbs faster than
acceptance) — the speedup here is bounded by the MoE, not the drafter, whose acceptance is strong.

## License & attribution

Original drafter © NVIDIA, released under the
[OpenMDW-1.1](https://openmdw.ai/license/1-1/) model license (see `LICENSE`). This repository
redistributes a dequantized (NVFP4 → bf16) copy under the same terms. Produced with
[NVIDIA Model Optimizer](https://github.com/NVIDIA/Model-Optimizer) (source) and
[mlx-dspark](https://github.com/ARahim3/mlx-dspark) (conversion).