Downloads · 30 days
52
16% of all-time downloads
DeepSeek-V4-Flash-0731-FP8
Pragmatism0220/DeepSeek-V4-Flash-0731-FP8
DeepSeek-V4-Flash-0731-FP8 is a machine learning model from Pragmatism0220. Use it for the machine learning task on the model card, and read the license before you ship it in a product. It is set up for sglang. The card lists the license as mit.
FP8 (block-wise e4m3 + f32 scale, 128×128 block) repackaging of deepseek-ai/DeepSeek-V4-Flash-0731.
All-time downloads
333
Public
Parameters
304B
307 GB on disk
Likes
2
Public
Repo makeup
Click a slice to open those files.
.safetensors307 GB · 100%
Parameter types
How the weights are stored.
F8_E4M3301B · 99%
Base models
Model card
From the Hugging Face model README
DeepSeek-V4-Flash-FP8
FP8 (block-wise e4m3 + f32 scale, 128×128 block) repackaging of deepseek-ai/DeepSeek-V4-Flash-0731.
What this is
The source checkpoint mixes three storage formats:
- MXFP4 (FP4 packed in int8 + E8M0 scale) — routed experts only
- MXFP8 (F8_E4M3 + E8M0 scale) — attention, shared experts, MTP
- BF16 / F32 — norms, embeddings, gates
This repo unifies everything into a single block-wise FP8 layout so the model can run through one FP8 kernel path.
Conversion
| Tensor | Source | Target |
|---|---|---|
ffn.experts.*.{w1,w2,w3} | int8 (FP4) + E8M0 scale | F8_E4M3 + F32 scale |
attn.*, shared_experts.*, mtp.*.e_proj | F8_E4M3 + E8M0 scale | F8_E4M3 + F32 scale (weight unchanged) |
attn.wo_a.weight | F8_E4M3 + E8M0 scale | BF16 (scale absorbed) |
attn.wo_a.scale | E8M0 | dropped |
| norms, embed, gate, etc. | BF16 / F32 | unchanged |
- Library
- sglang
- Type
- deepseek_v4
- License
- mit
- Languages
- en, zh
- Created
- Aug 2, 2026
- Updated
- Aug 3, 2026