Downloads · 30 days
24
26% of all-time downloads
deepseek-v4-flash-dspark-xiaoyi-v0.1
xpzhao11/deepseek-v4-flash-dspark-xiaoyi-v0.1
deepseek-v4-flash-dspark-xiaoyi-v0.1 is a machine learning model from xpzhao11. Use it for the machine learning task on the model card, and read the license before you ship it in a product.
This repository contains only the trained DSpark draft parameters (mtp.).
All-time downloads
94
Public
Repo size
39.7 GB
Likes
0
Public
Repo makeup
Click a slice to open those files.
.safetensors39.7 GB · 100%
At a glance
- Model type
- deepseek_v4
- Access
- Public
- Created
- Jul 27, 2026
- Updated
- Jul 27, 2026
- SHA
- 0462a97a
Model card
From the Hugging Face model README
DeepSeek-V4 Flash DSpark Draft
This repository contains only the trained DSpark draft parameters (mtp.*).
Contents
- Base model:
DeepSeek/DeepSeek-V4-Flash-DSpark - Draft tensor prefix:
mtp. - Storage dtype:
bfloat16 - Draft tensors: 2376
- Draft shards: 8
- Target weights: not included
- Quantization scales: not included
Usage
This is a draft-only training artifact, not a standalone Hugging Face model.
It cannot be loaded directly using AutoModel.from_pretrained() or served
directly with vllm serve.
Users must combine these mtp.* tensors with the base Target model and
quantize the Draft according to their inference backend.
Parameter-to-shard mappings are stored in
dspark-draft.safetensors.index.json.
- Type
- deepseek_v4
- Created
- Jul 27, 2026
- Updated
- Jul 27, 2026