Run DeepSeek on your own server
Ollama, llama.cpp, vLLM, SGLang, LM Studio, quantization
172 projects, 89 active.
| Project | Stars | Status | Last push | Description / install |
|---|---|---|---|---|
| TranslateBooksWithLLMs | 2,569 | active | 2026-11-03 | Translate full-length books and documents with Ollama, OpenAI-compatible, Gemini, Mistral, DeepSeek, Poe or Op |
| vLLM-Moet | 567 | active | 2026-11-04 | Frontier MoE models on RTX PRO 6000 / RTX 5090 (sm_120): DeepSeek-V4.1-Flash on 4x RTX PRO 6000, as a GPU+CPU |
| PureChat | 545 | active | 2026-09-17 | PureChat 是一款全面的聊天应用程序，它将传统的即时消息与高级 AI 功能集成在一起。其模块化架构、可扩展的 AI 集成系统和灵活的配置选项使其成为强大的应用程序，也是 AI 增强通信应用程序的坚实基础。支持Ope |
| vllm-deepseek-v4-sm89 | 202 | active | 2026-11-03 | Run DeepSeek-V4.1-Flash / DeepSeek-V4-Flash and GLM-5.3-Flash on SM89 (Ada / RTX 4090) and SM120 (RTX PRO 6000 |
| DeepSeek-V4.1-Flash-TP2-and-TP4-DGX-Sparks | 159 | active | 2026-10-26 | DeepSeek-V4.1-Flash-1017 on NVIDIA DGX Sparks with TensorFold + EXL3: TP4 (4 Sparks, 93.6 tok/s, 1M context) a |
| pd-bridge | 151 | active | 2026-10-16 | Heterogeneous prefill/decode for DeepSeek-V4-Flash: CUDA prefill (DGX Spark, vLLM) -> Metal decode (Mac Studio |
| deepseek-v41-tensorfold-spark | 127 | active | 2026-11-01 | DeepSeek-V4.1-Flash on 2x NVIDIA DGX Spark with the TensorFold engine: 1.8-1.9x code, 2.5x multi-stream, 1.7-1 |
| chatgpt-shell thin readme | 1,210 | active | 2026-09-18 | A multi-llm Emacs shell (ChatGPT, Claude, DeepSeek, Gemini, Kagi, Ollama, Perplexity) + editing integrations |
| local_ai_ocr slow >2m | 887 | slow | 2026-07-05 | An local, offline (after initial setup), portable OCR software that can process images and PDF files, using De |
| twinspark slow >2m | 176 | slow | 2026-08-14 | Production-grade recipe for DeepSeek-V4-Flash-0731 (284B MoE) on 2x NVIDIA DGX Spark: self-healing 2-node vLLM |
| deepseek-v4-cmp170hx slow >2m | 118 | slow | 2026-08-13 | DeepSeek-V4-Flash-0731 on 4x CMP 170HX (sm_80): 98 tok/s decode, ~5300 tok/s prefill. DSpark speculative decod |
| DeepSeek-V4.1-Flash-4x-DGX-Spark-TP4 | 87 | active | 2026-11-02 | Tuned TP4 serving profile for DeepSeek-V4.1-Flash on 4x NVIDIA DGX Spark (SGLang, DSpark), built on the MiaAI- |
| spark-ds41f | 85 | active | 2026-11-04 | Reproducible Docker/vLLM deployment, tuning, and benchmarks for DeepSeek V4.1 Flash on a switchless three-node |
| dspark-vllm-gx10 | 83 | active | 2026-10-27 | Two-node DGX Spark/ASUS GX10 DeepSeek V4 Flash DSpark NVFP4 port for vLLM 0.25, with live dashboard and reprod |
| DeepSeek-V4.1-Flash-1017-EXL3-TensorFold-2x-DGX-Spark | 76 | active | 2026-11-01 | DeepSeek-V4.1-Flash-1017 EXL3 2.9bpw on 2x NVIDIA DGX Spark (GB10) with TensorFold: 2.1-2.3x the vLLM kit's de |
| deepseek-v4.1-flash-ascend910B | 57 | active | 2026-11-04 | DeepSeek-V4.1-Flash W4A8 optimized vLLM deployment for Ascend 910B (A2) / 910C (A3): patches, one-command serv |
| deepseek-colab | 49 | active | 2026-10-30 | Uncensored DeepSeek-V4.1-Flash IQ2_XS on Google Colab |
| dgx-spark-deepseek-v41 | 46 | active | 2026-10-22 | DeepSeek-V4.1-Flash on two NVIDIA DGX Sparks: EXL3 weights from the official checkpoint, upstream vLLM with ou |
| ComfyUI-LLM-Session | 37 | active | 2026-10-08 | Local LLM session nodes for ComfyUI that run GGUF models with llama.cpp without external runtimes like Ollama. |
| DeepSeek-V4.1-Flash-1017-FP8-abliterated-4x-RTX-PRO-6000 | 35 | active | 2026-10-28 | Abliterated DeepSeek-V4.1-Flash-1017 (FP8 dense, FP4 experts) on 4x RTX PRO 6000 with vLLM-Moet r21 and DSpark |
| sglang-sxm2 | 34 | active | 2026-10-07 | Agentic-coding oriented inference engine for N x V100-SXM2, based on SGLang: GLM-5.3-Flash, Qwen3.8-Flash-Next |
| LuZ-0.1.7-DeepSeek-v4.1-Flash-DGXspark-TP4-Ring | 31 | active | 2026-10-01 | DeepSeek-V4.1-Flash on 4x DGX Spark TP4 switchless RoCE ring - SGLang production recipe, 600K context, DSpark |
| DeepSeek-V4.1-Flash-EXL3-vLLM-2x-DGX-Spark | 28 | active | 2026-09-27 | DeepSeek-V4.1-Flash EXL3 vLLM recipe for 2x DGX Spark (GB10) |
| ds4-vllm | 26 | active | 2026-09-24 | A Deepseek V4 Flash 0731 / Vision Exp VLLM stack |
| dsv41-flash-cmp170hx-recipe | 20 | active | 2026-11-02 | DeepSeek-V4.1-Flash on 4× NVIDIA CMP 170HX with vLLM: up to 297 tok/s for one user, up to 868 tok/s at eight, |
| deepseek-v4.1-flash-a100-turbo | 20 | active | 2026-09-15 | DeepSeek-V4.1-Flash inference on NVIDIA Ampere with vLLM, SM80 patches, reproducible throughput and coding-age |
| claude-code-delegate-local | 19 | active | 2026-11-02 | MCP server that delegates Claude Code subagents to local models (LiteLLM/llama.cpp/Ollama), DeepSeek, AWS Bedr |
| a4agent | 18 | active | 2026-10-30 | a4agent（0.4.0 起由 a4api 是「八端 AI 编程工具管理台 + 本地大模型推理控制台」：为 Claude Code、Codex、dsh（DeepSeek Harness）、ZCode、pi、Qoder、 |
| deepseek-v41-flash-on-5090 | 18 | active | 2026-09-23 | DeepSeek-V4.1-Flash (552B MoE) on one RTX 5090 with a llama.cpp fork: report, tools, results |
| SnowLLM | 18 | active | 2026-11-02 | Your Ryzen AI Max is running at half speed. SnowLLM: up to 9x faster prefill and 2x faster decode than llama.c |
| vLLM-Moet-GB10 | 16 | active | 2026-10-24 | Run DeepSeek-V4-Flash-0731, GLM-5.3-Flash on a single NVIDIA GB10 / DGX Spark - 2-bit MoE planes, FP4 quality |
| machx-inference-engine | 10 | active | 2026-11-03 | C++/SYCL local LLM inference for Intel Arc: Qwen3.8, DeepSeek-V4.1, MiMo-V2.6, GLM-5.3 and LTX-2.5 text-to-vid |
| dsh-force-compact | 10 | active | 2026-10-30 | Aggressive context compaction for local-first agents. Runs Qwen3.8‑27B on self‑hosted llama.cpp at low context |
| glm-5.2-sm120 slow >2m, no license | 155 | slow | 2026-06-19 | GLM-5.2-NVFP4-REAP-469B serving on SM120 (4× RTX PRO 6000 Blackwell) — one-command vLLM launch recipe, 250K co |
| vllm-ascend-deepseek-v41-a2 no license | 19 | active | 2026-11-02 | DeepSeek-V4.1-Flash 在 8×Ascend 910B3 (A2) 上的 vllm-ascend v0.30.0rc2 调优补丁与起服脚本 |
| deepseek-v4-sm89 | 8 | active | 2026-11-02 | 面向 8×RTX 4090（SM89）的 DeepSeek-V4-Flash-0731 MXFP4 SGLang 实验分支 |
| dsv41-flash-rtx-pro-6000 | 7 | active | 2026-09-11 | DeepSeek-V4.1-Flash on four RTX PRO 6000 Blackwell (sm_120): the SGLang recipe, two patches, 11 tuned block-FP |
| lingualens | 7 | active | 2026-10-26 | Translate anything on the web — with local LLMs (Ollama, LM Studio), cloud APIs (OpenAI, DeepSeek, DeepL, Goog |
| DeepSeek-V4.1-Flash-4x-GB10-1M-Full-Recipe | 6 | active | 2026-09-28 | Measured serving recipes for DeepSeek-V4.1-Flash on 4x NVIDIA DGX Spark (GB10): 1M context on vLLM (CUDA graph |
| DeepSeek-V4.1-Flash-GGUF-DGX-Spark-recipe | 6 | active | 2026-09-11 | llama.cpp recipe: convert DeepSeek-V4.1-Flash (deepseek41 arch, engram) to GGUF on one DGX Spark GB10. Convers |
| dsh-plugin-local-model | 6 | active | 2026-09-29 | "DeepSeek Harness 本地模型插件：设置里独立的「本地模型」页，首条对话自动拉起 llama.cpp，空闲 5 分钟自动卸载释放资源。" |
| ds41-flash-hybrid | 6 | active | 2026-10-18 | DeepSeek-V4.1-Flash hybrid CPU/GPU serving recipe (KTransformers + SGLang) for a single 96 GB workstation GPU |
| flyweight | 6 | active | 2026-11-04 | A native C++/CUDA GGUF inference runtime with an OpenAI- and Anthropic-compatible server. Runs Qwen, DeepSeek- |
| sglang-deepseek-v41-flash-sm120 | 5 | active | 2026-10-30 | Reproducible SGLang image for DeepSeek-V4.1-Flash on four RTX PRO 6000 Blackwell GPUs (SM120), TP4 |
| dsh-local-model | 5 | active | 2026-10-14 | 让 DeepSeek Harness 直连本地模型：兼容 OpenAI 协议（llama.cpp / LM Studio / vLLM / Ollama 的 /v1）与 Ollama 原生 API。无需登录、不消耗账号 |
| deepseek-v4.1-flash-1017-ktransformers | 4 | active | 2026-11-04 | Run DeepSeek-V4.1-Flash-1017 with SGLang + KTransformers (CPU/GPU hybrid): one setup script, a launcher that r |
| superl8-serve | 4 | active | 2026-09-23 | OpenAI-compatible INT8 inference server for CMP 100-210 and Volta GPUs. DP4A kernels, native GGUF loading, spe |
| dsh-llamacpp-bridge | 4 | active | 2026-10-22 | DeepSeek Harness plugin: local llama.cpp (llama-server) as a first-class model provider — process/router manag |
| inferhost | 4 | active | 2026-09-14 | Self-hosted multi-modal AI server: chat/vision LLMs (Qwen, Llama, Gemma, DeepSeek), text-to-speech (OuteTTS, Q |
| deepseek-v4-flash-v100 | 3 | active | 2026-11-03 | DeepSeek-V4-Flash and V4.1-Flash on NVIDIA V100: llama.cpp recipe and quantized-KV fixes (V4); 1Cat-vLLM PP2xT |
| DeepSeek-V4.1-Flash-SGLang-DGX-Spark | 3 | active | 2026-09-11 | Measured DeepSeek-V4.1-Flash SGLang recipe for 4/8 DGX Spark nodes |
| plump_fish-build-kit | 3 | active | 2026-10-19 | Windows turnkey installer that attaches a local Qwen3-VL vision model (llama.cpp) to DeepSeek Harness: CUDA/Vu |
| deepseek-v41f1017-gb10-sglang-tp8 | 2 | active | 2026-10-27 | 8x GB10 (DGX Spark + 超聚变) DeepSeek-V4.1-Flash-1017：SGLang TP8 / EP8，1M 窗 / 100K 输出 / KV 8M；0到1部署手册 + 固定起服链 + 内 |
| helioslm | 2 | active | 2026-11-04 | Pure-PyTorch DeepSeek-V3/K3-style LLM stack where every serving optimization ships with a bit-exactness oracle |
| Siege-Ladder | 2 | active | 2026-11-04 | DeepSeek-V4.1-Flash (1017) on one 16 GB GPU + 64 GB RAM, experts streamed from NVMe: llama.cpp patches, a weig |
| traliran-ai-hub | 2 | active | 2026-11-03 | A lightweight, serverless, and privacy-first AI web client & playground with a built-in code execution sandbox |
| sglang-deepseek-v41-flash-sm120-compat | 2 | active | 2026-11-04 | DeepSeek-V4.1-Flash-1017 compatibility fixes and validated TP4/EP2 serving profile for RTX PRO 6000 Blackwell |
| nongor | 2 | active | 2026-10-02 | Local-first RAG chat that answers questions about your documents with citations. BGE-M3 embeddings, hybrid den |
| DeepSeek-V4.1-Flash-1017-UNCENSORED-DGX-Sparks | 1 | active | 2026-11-01 | Switchless TP4/TP2 for DeepSeek-V4.1-Flash-1017 (uncensored) on DGX Spark (GB10). TensorFold Zig TP4 on a swit |
| deepseek-v4-flash-0731-sglang-4x-rtx-pro-6000 | 1 | active | 2026-10-13 | Run DeepSeek-V4-Flash-0731 on 4× RTX PRO 6000 with patched SGLang, fixing DSPARK crashes and draft-depth corru |
| deepseek-v4.1-flash-dspark-tp4-4xgb10 | 1 | active | 2026-09-14 | BertholomusAI in-house recipe: DeepSeek-V4.1-Flash on 4x NVIDIA DGX Spark (GB10), TP4 DSpark under SGLang - co |
| dsa-multi-layer-indexer | 1 | active | 2026-11-04 | Multi-layer Indexer (MLI): hide the DeepSeek Sparse Attention indexer behind attention in DeepSeek-V3.2 and GL |
| dsv4-vision-sm120 | 1 | active | 2026-09-14 | DeepSeek V4 Flash Vision on 2× RTX PRO 6000: vision at zero cost over 0731. Patches for FlashInfer SM120 spars |
| blackwell-serving-kit | 1 | active | 2026-10-21 | Launch recipes, measured tokens/s and the traps behind them for serving Qwen3.8-Flash-Next, DeepSeek-V4-Flash, |
| quantserve | 1 | active | 2026-10-29 | 开源大模型本地化部署与量化推理平台 — Qwen3.8-2.4T-A95B / DeepSeek V4 Pro 等开放权重旗舰模型的下载、量化、MoE并行规划与多引擎推理一站式工具（零依赖核心） |
| dsh-localforge | 1 | active | 2026-11-03 | Connects DSH to local AI models |
| deepseek-v4.1-flash-reap-thor | 1 | active | 2026-10-31 | DeepSeek-V4.1-Flash-1017 pruned to 192 of 384 experts per layer, at ~2.85 bpw with Engram read from NVMe, serv |
| deepseek-v4-flash-mac | 1 | active | 2026-10-26 | Run DeepSeek-V4.1-Flash (551B MoE, IQ2_XXS) locally on Apple Silicon with 128K context, even though it's bigge |
| MacSeek | 1 | active | 2026-11-03 | From-scratch Metal inference engine that runs DeepSeek-V4.1-Flash (233 GB GGUF) with 200K context on a 64 GB M |
| sami-deepseek-moe | 1 | active | 2026-10-18 | Hardware-decoupled local inference suite for DeepSeek-MoE-16B-Chat via llama.cpp (-cmoe) and SAMI-AsymQ. |
| llamapad-dsh-plugin | 1 | active | 2026-10-17 | DeepSeek Harness (DSH) plugin for llama.cpp docker panel "llamapad" / llamapad 面板的 DeepSeek Harness 插件 |
| dsh-plugin-local-infer | 1 | active | 2026-11-04 | A DeepSeek Harness plugin to install and manage local inference providers like Strata, Ninfer, VLLM, llama.cpp |
| top-open-weight-coding-models | 1 | active | 2026-11-02 | Open-weight coding and agent models (Qwen Coder, DeepSeek, GLM, Kimi, Devstral, StarCoder) plus the tools to r |
| Strata-JEV | 1 | active | 2026-10-08 | Strata 125B MoE + JEV mode: llama.cpp fork with parallel-decision endpoints (yes/no in ~10ms), an MCP server, |
| DeepSeek-V4-Flash-0731-CRACK-2x-DGX-Spark slow >2m | 91 | slow | 2026-08-14 | Reproducible refusal-subspace editing and TP=2 deployment of DeepSeek V4 Flash 0731 on two DGX Sparks |
| Deepseek-v4-Flash-TP2-DGX-Spark-500k-CTX slow >2m | 85 | slow | 2026-07-13 | Working recipe to serve DeepSeek-V4-Flash across two NVIDIA DGX Spark (GB10) nodes with vLLM (TP=2, FP8 KV, MT |
| sglang-deepseek-v4-flash-sm120 slow >2m | 37 | slow | 2026-08-25 | Reproducible SGLang image for DeepSeek-V4-Flash-0731 on SM120 |
| DeepSeek-V4-Flash-DSpark-v026-SM121 slow >2m | 31 | slow | 2026-08-03 | DeepSeek-V4-Flash-DSpark optimized vLLM 0.26.0 SM121 dual-GB10 evidence (NVFP4 KV, B12X, DSpark K6) |
| deepseek-v4-flash-0731-sglang-4x-rtx-pro-6000 slow >2m | 21 | slow | 2026-08-09 | Reproducible SGLang recipe + public prebuilt image (ghcr.io) for DeepSeek-V4-Flash-0731 on 4x RTX PRO 6000 Bla |
| ds4-spark-vllm slow >2m | 18 | slow | 2026-05-11 | antirez/ds4-style hybrid quant DeepSeek V4 Flash on a single DGX Spark via vLLM |
| ai-cli-switchboard slow >2m | 15 | slow | 2026-07-07 | A simple framework for using Claude Code or Codex CLI as the frontend to any cloud or local LLM on Apple Silic |
| deepseek-v4-a100-sglang-v0516 slow >2m | 14 | slow | 2026-08-02 | 在四卡A100上适配Sglang部署deepseek-v4-flash正式版模型 |
| sglang-v0522-sm120-dsv4.1 no license | 5 | active | 2026-10-29 | SGLang v0.5.22 — DeepSeek V4.1 Flash on RTX PRO 6000 (SM120) patch |
| r820-llm-public no license | 4 | active | 2026-11-03 | DeepSeek V4.1 Flash on a 2012 Dell PowerEdge R820: ik_llama.cpp patch and tools |
| LocalAI-Package no license | 2 | active | 2026-10-28 | On-device Small Language Model package for Automation Anywhere A360 - CPU inference via llama.cpp with Qwen 2. |
| dsv41-4x-spark-tutorial no license | 2 | active | 2026-09-17 | DeepSeek-V4.1-Flash 4× DGX Spark 环形 RoCE 集群部署教程 (SGLang TP4 / switchless ring) |
| agents-on-modal no license | 2 | active | 2026-11-04 | Run Claude Code and Codex agents (solo or in 100-agent sandboxed swarms) against self-hosted open-weight and a |
| cuda-sm120-deepseek-course no license | 1 | active | 2026-10-21 | Blackwell (sm_120) 架构大模型推理 CUDA & CUTLASS 算子优化实战课程（针对 DeepSeek-V4.1-Flash / vLLM / SGLang） |
| Vibe_Engine no license | 1 | active | 2026-10-08 | Deterministic local execution runtime for any LLM that can use MCP + REST for Claude, Antigravity, Gemini, Git |
| LLM-Finetuning | — | active | 2026-10-25 | Reasoning distillation with QLoRA: fine tuning Llama 3.2 3B on reasoning traces distilled from DeepSeek R1 usi |
| deepseek-v4-gguf | — | active | 2026-09-29 | (IQ1_S-XL) quantized in 2 shards [50GB + 11.6GB] |
| deepseek-v41-upstream | — | active | 2026-11-03 | Run ISTA-DASLab's DeepSeek-V4.1-Flash GSQ-RCO quant (748B) on stock upstream llama.cpp — repack tool + one-GPU |
| purple-squirrel-r1 | — | active | 2026-09-08 | 8B reasoning LLM fine-tuned with DeepSeek-R1 + MLX LoRA on Apple Silicon. GGUF quantizations, Ollama Modelfile |
| dsh-plugin-lmstudio-image-fix | — | active | 2026-10-28 | Плагин для DeepSeek Harness (DSH), который убирает ошибку LM Studio в работе с изображениями |
| dsh-local-llm | — | active | 2026-10-31 | Detect, probe and register local OpenAI-compatible LLMs (LM Studio, Ollama, vLLM, llama.cpp) as DeepSeek Harne |
| vllm-dsv4-ampere slow >2m, no license | 35 | slow | 2026-07-02 | DeepSeek-V4-Flash on Ampere SM 8.6 via vLLM (pyref kernel replacements) |
| deepseek-v4-local-deploy slow >2m, no license | 26 | slow | 2026-05-16 | DeepSeek V4 Pro/Flash 本地部署指南汇总：vLLM、SGLang、DwarfStar4（Metal）、双DGX Spark混合后端 |
| LuZ0.3.1-DeepSeek-v4-Flash-DSpark-DGXspark-TP4-Ring slow >2m, no license | 16 | slow | 2026-08-26 | W4A4 full（VLLM_MOE_W4A4=2）+ （VLLM_B12X_SHARED_WRAPPER=1）+ FlashInfer 0.6.16 + threshold 4096 + util 0.78 + DSp |
| DeepSeek-FineTuner slow >2m, no license | 12 | slow | 2026-08-10 | A quantized GGUF reasoning model optimized for local, offline inference with Ollama. |
| dsv4-on-ampere slow >2m, no license | 11 | slow | 2026-08-07 | Serving DeepSeek-V4-Flash on Ampere (A100/A800, SM80) with SGLang — stock checkpoint, one four-line gate |
| llama.cpp-deepseek-v4 slow >2m | 10 | slow | 2026-05-12 | |
| deepseek-v4-flash-fp8-repack slow >2m | 9 | slow | 2026-08-06 | Fine-tune DeepSeek-V4-Flash on H100s, then serve it — repack BF16 merges into the official blockwise-FP8 layou |
| deepseek-v4-sm80 slow >2m | 6 | slow | 2026-08-18 | Optimized DeepSeek V4 Flash inference and serving on NVIDIA Ampere GPUs (SM80), based on Mini-SGLang, with TP8 |
| lmstudio-dsv4-patch slow >2m | 6 | slow | 2026-05-27 | Sideload antirez's DeepSeek V4 Flash llama.cpp fork into LM Studio (Mac, 128GB unified memory). Patches for AB |
| DeepSeek-V4-Flash-Vision-SGLang-DGX-Spark slow >2m | 4 | slow | 2026-09-04 | DeepSeek-V4-Flash-Vision-Exp on NVIDIA DGX Spark with SGLang: TP2 (2 Sparks) live with real-prompt benchmarks |
| llama_cpp_for_radxa_dragon_wing_q6a slow >2m | 4 | slow | 2026-07-24 | Modifiers to get Llama.cpp working using the built in NPU on the Radxa Dragon Wing Q6A SBC (Qualcomm QCS6490 c |
| deepseek-harness-lmstudio slow >2m | 4 | slow | 2026-08-14 | Configure DeepSeek Harness for LM Studio and verify Qwen tool calling end to end. |
| deepseek-vl-support slow >2m | 3 | slow | 2026-09-01 | Give DeepSeek (text-only) models **vision** in Claude Code and Codex by routing image files to any OpenAI-comp |
| local-llm-mcp slow >2m | 3 | slow | 2026-07-01 | MCP server that lets Claude Code & Codex delegate bounded, token-heavy generation to local or cheap-cloud LLMs |
| codex_proxy_for_all_models slow >2m | 3 | slow | 2026-07-05 | Lightweight OpenAI-compatible proxy that lets Codex route Responses API traffic to NVIDIA Build, Ollama, LM St |
| dsv4-two-node-strix-halo slow >2m | 3 | slow | 2026-08-05 | DeepSeek-V4-Flash across two Strix Halo boxes over 100GbE RDMA: llama.cpp patches, launch config, and measured |
| LLM-Bridge-for-Copilot-Chat slow >2m | 2 | slow | 2026-08-15 | 将 Copilot Chat 路由到任意 OpenAI 兼容端点：DeepSeek、MiMo、OpenCode Zen / Go、OpenRouter，以及本地模型（llama.cpp / Ollama / LM Stu |
| forgequant slow >2m | 2 | slow | 2026-06-19 | Config-driven asymmetric quantization for DeepSeek-V4-Flash: a recipe (per-family quant types + imatrix) -> a |
| open-llm-benchmarks slow >2m | 2 | slow | 2026-06-18 | Open-source LLM inference benchmarks — TTFT, TPOT, Throughput, Latency & Cost-per-token for models like Llama, |
| VibeChat slow >2m | 2 | slow | 2026-07-30 | A sleek, ultra-lightweight hybrid LLM chat client built with Next.js (App Router) and Tailwind CSS. Seamlessly |
| dsh-vision-LMstudio slow >2m | 1 | slow | 2026-08-19 | 让你能通过deepseek harness调用LM studio加载的本地视觉模型 |
| PDF-Translator slow >2m | 1 | slow | 2026-08-08 | Translate PDFs per-page on demand with a side-by-side bilingual view. Works with free Google Translate, local |
| deepseek-v4-flash-8x4090 slow >2m | 1 | slow | 2026-08-17 | DeepSeek-V4-Flash-0731 deployment and benchmarks on 8×RTX 4090 24GB, covering ds4, vLLM, SGLang, GGUF, DSpark |
| Unlimited-OCR-WebUI slow >2m | 1 | slow | 2026-07-22 | High-performance local WebUI for Advanced Document OCR and Layout Analysis. Powered by DeepSeek-OCR / DeepSeek |
| dsh-llama-responses slow >2m | 1 | slow | 2026-08-24 | DeepSeek Harness plugin: run subagents on a local llama.cpp model via the OpenAI Responses (/v1/responses) pro |
Verified 2026-11-04. GitHub API + raw README.