DeepSeek

DeepSeek V4 Flash

Name: DeepSeek V4 Flash
Rating: 96 (8 reviews)
Author: DeepSeek

前沿

HuggingFace

3.4M下载量1.3K点赞Apr 2026发布日期1.0M tokens上下文MIT许可证98 卓越质量

DeepSeek V4 Flash (284B parameters) requires approximately 160.8 GB of VRAM with NVFP4 quantization. As a Mixture of Experts model with 13B active parameters, it uses less memory than its total parameter count suggests. For the best balance of quality and speed, we recommend hardware with at least 185 GB of VRAM.

快速开始

— 复制粘贴即可本地运行

Copy-paste commands to run DeepSeek V4 Flash on your machine.

Run

docker run --rm -it ghcr.io/ggerganov/llama.cpp:full \
  --hf-repo "deepseek-ai/DeepSeek-V4-Flash" \
  --hf-file "DeepSeek-V4-Flash-NVFP4.gguf" \
  -c 4096 -ngl 99

Quick specs

Parameters284B (13B active)

Architecturemoe (MoE)

Context1.0M tokens

Modalitytext

Min RAM110.8 GB

Rec. RAM159 GB (NVFP4)

LicenseMIT

FamilyDeepSeek

✓ Code✓ Reasoning

About this model

DeepSeek V4 Flash is the lighter 284B-parameter sparse MoE sibling of V4 Pro (13B active, 256 routed + 1 shared expert) with the same 1M-token context. Experts ship natively in FP4, so the real on-disk footprint is roughly 158 GB rather than the FP16 size — it fits a single 192 GB unified-memory machine or a 2-4 GPU server while keeping near-frontier reasoning and coding quality.

•284B total / 13B active sparse MoE — 256 routed + 1 shared expert
•Native FP4 experts: ~158 GB on disk
•1M-token context with near-frontier coding quality
•Runs on a single 192 GB unified-memory box or a small GPU server

DeepSeek V4 Flash 的最佳选择

AMD Instinct MI325X 256GBS

256 GB

AMD Instinct MI350X 288GBS

运行此模型

DeepSeek V4 Flash on AMD Instinct MI325X 256GB DeepSeek V4 Flash on AMD Instinct MI350X 288GB DeepSeek V4 Flash on NVIDIA B200 180GB

量化选项

各量化级别的 VRAM 估算

No hardware detected — fit column shows raw VRAM estimates

Quant	Bits	VRAM	Quality	Fit
Q2_K	2	110.8 GB	Low	—
Q3_K_S	3	139.2 GB	Low	—
NVFP4	4	159.0 GB	Medium	—
Q4_K_M	4	173.2 GB	Medium	—
Q5_K_M	5	204.5 GB	High	—
Q6_K	6	232.9 GB	High	—
Q8_0	8	303.9 GB	Very High	—
F16	16	582.2 GB	Maximum	—

Quality benchmarks

DeepSeek V4 Flash benchmark scores

Benchmark verified

Coding

SWE-bench Verified—

HumanEval+—

Aider Polyglot—

LiveCodeBench91.6%

Reasoning

MMLU-Pro86.2%

GPQA Diamond—

MATH-500—

ARC Challenge—

Source: vendor-reported · 2026-04-24

硬件兼容性

全部硬件的适配估算

打开计算器

Computing compatibility...

内存详细分析

Reference: RTX 2060 6GB

Weights158.0 GB

KV Cache1.3 GB

Runtime0.9 GB

Headroom0.6 GB

常见问题

FAQ — DeepSeek V4 Flash

另请参阅

量化指南评分方法打开计算器