deepseek-v4-flash 开源项目推荐（44 个）

NVIDIA-NeMo /Automodel

🚀Pytorch用于LLC/VLM的分布式本地培训库，具有OOTB Hugging Face支持

notatestuser /ds4-control

ds4.c（经过微调的DeepSeek V4 Pro/Flash）的macOS菜单栏前端，具有1 M上下文。

deepseek-php /deepseek-php-client

⚡️ 一个由社区驱动的 DeepSeek AI PHP 客户端，旨在为 PHP 应用带来干净的 API 访问、流畅的开发者体验和框架友好的集成。

yxlao /deepseek-cursor-proxy

修复 DeepSeek V4 缺失的reasoning_content

Entrpi /ds4-on-spark

antirez/ds4（DwarfStar 4）在NVIDIA DGX Spark上——安装、基准测试和车顶线分析。在带宽上限的~95%处实现稳态解码;MTP与并发分析。

snake-aabb-wtf /deepseek-web2api-free

基于 FastAPI 的 DeepSeek Chat 反向代理，将 DeepSeek 网页版的 API 转换为 OpenAI 兼容格式。 支持流式/非流式对话、专家模式、深度思考（reasoning_content）、工具调用（DSML prompt injection）。 自动处理 PoW 鉴权挑战，无需官方 API Key。

yanun0323 /Whallm

DeepSeek-V4-Flash-0731 284B inference in ~25 GB of RAM / Qwen3.8-Next-Flash inference in ~15 GB of RAM on any M-series MacBook

baker27727 /ds4-8gb-cpu

DeepSeek V4 Flash CPU/NVMe research fork: 78.62 GiB GGUF validated on 7.7 GiB RAM, CPU-only, using demand paging.

local-ai-zone /local-ai-zone.github.io

My public AI learning journey — sharing what I learn, build, test, and discover across AI. Traffic: 7K+ monthly clicks · 150K+ AI citations/month · 5.5 average search position across search engines. local-ai-zone.github.io

pepuscz /strix-halo-llm

Winning Qwen and DeepSeek setups for AMD Strix Halo: pinned Ansible recipes and matched benchmarks on a 128 GiB Ryzen AI Max+ 395 / Radeon 8060S.

HELPMEEADICE /deepseek2api-neo

DeepSeek 逆向 API 支持 Deepseek V4

flaqai /deepeseek-harness-guide

Guide for development with DeepSeek Harness. Building plugin for DeepSeek Harness Project.

Reederey87 /dgx-spark-2x-deepseek-v4-flash

Reproducible kit to deploy DeepSeek-V4-Flash-DSpark on a 2× NVIDIA DGX Spark (GB10) cluster: vLLM TP=2 over QSFP 200GbE, NVFP4 KV, DSpark speculative decoding, 1M context, systemd self-heal. Apache-2.0.

NimbleMarkets /ds4go

该项目在 GitHub 上没有填写描述

QubridAI-Inc /qubrid-cookbook

A collection of recipes/notebooks showcasing use-cases of open-source models with Qubrid AI.

SIMON-WORLD /codex-deepseek-vision

Codex vision bridge for DeepSeek V4 Flash: give text-only DeepSeek image capability in Codex. Local proxy turns pasted images and view_image into text via free GLM-4V-Flash or any OpenAI-compatible vision API. No GPU, no Ollama.

yuyukosama2004 /codex-deepseek-max-guide

Codex 桌面版为 DeepSeek-V4-Flash 开启 Max 推理档位的完整排障与配置指南 / Complete guide to enable Max reasoning effort for DeepSeek-V4-Flash in Codex desktop

LimeChain /vllm

DeepSeek-V4-Flash-0731 on 8x RTX 3090 (SM86) with vLLM — verified FP8 serving, benchmarks, build guide, and reproducible release.

mrbeandev /deepseek-browser-agent

Supercharge your daily workflows! Autonomously browse pages, scrape content, switch tabs, fill out inputs, and stream Chain-of-Thought (CoT) reasoning under granular human-in-the-loop safety switches and glassmorphic UI controls.

dyrnq /ai-adapter

A Rust proxy that exposes an **OpenAI Responses API** interface, translating to various upstream backends (DeepSeek, OpenAI, Anthropic) with full streaming SSE support. Built for Codex CLI and similar tools hardcoded to OpenAI.