Локальные LLM
Qwen3-Coder
Qwen3-Coder - Alibaba's open code model series for agentic coding, up to 1M tokens of context
★ 16,8 тыс. звёздNOASSERTION
Смотреть руководство →DeepSeek-Coder-V2 — проект с открытым исходным кодом из категории «Локальные LLM» от deepseek-ai (deepseek-ai/DeepSeek-Coder-V2 на GitHub). У него 7 014 звёзд на GitHub и 1 108 форков, он распространяется по лицензии MIT и последний раз обновлялся на GitHub 11 ноября 2025.
DeepSeek-Coder-V2 — DeepSeek's open code model family that challenged closed-source code models.
Choose the variant that fits your hardware — a Lite model for a single workstation GPU, or the full model only if you have a multi-GPU server — and your inference stack (Transformers or SGLang).
Clone deepseek-ai/DeepSeek-Coder-V2 from GitHub (or use the install path in the run steps below) and skim its README for the parts relevant to your goal.
There is nothing to build: download the weights from the model's Hugging Face page and load them with Transformers or SGLang using the README's examples.
Run the code-completion example with DeepSeek-Coder-V2-Lite-Base on a short Python snippet, then try the chat example with the Instruct variant.
Evaluated extensively on code generation benchmarks in the repo, positioning it against leading closed models.
Supports fill-in-the-middle style code insertion as well as plain completion, with worked examples for both.
Evaluated on code-fixing tasks, so it can be used to repair broken code, not just write new code.
Instruct (chat) and Base variants, in full and Lite sizes, cover assistant use and raw completion/fine-tuning bases.
The repo also reports mathematical-reasoning and general natural-language evaluations, beyond pure code tasks.
Run with Hugging Face Transformers for simplicity, or SGLang (the recommended path) for efficient serving.
Get the source from the official GitHub repository for DeepSeek-Coder-V2.
git clone https://github.com/deepseek-ai/DeepSeek-Coder-V2.git
cd DeepSeek-Coder-V2Use the README's Hugging Face example to load the Lite base model for code completion.
python -c "from transformers import AutoTokenizer, AutoModelForCausalLM; m='deepseek-ai/DeepSeek-Coder-V2-Lite-Base'; tok=AutoTokenizer.from_pretrained(m, trust_remote_code=True)" # full example in READMEFor the full model, follow the README's SGLang section (BF16, tensor parallelism) to serve it efficiently.
# See README 'Inference with SGLang (recommended)' for the exact serve command and settingsdeepseek-ai (DeepSeek), through the deepseek-ai/DeepSeek-Coder-V2 repository, with models on Hugging Face and an API platform documented in the repo.
Yes — the project is open source under the MIT license as recorded for this listing. You can use, modify and distribute it under the MIT terms.
Running locally needs no API key — just hardware. DeepSeek also offers a hosted API platform if you prefer not to run the weights yourself.
A hosted API is instant and scales for you, billed per token. DeepSeek-Coder-V2 gives you open weights to run, inspect and fine-tune on your own GPUs — full control and no per-token bill, with real hardware requirements, especially for the full model.
Clone the repository (or use the documented installer), follow the install and configure steps on this page, and run the smallest real task first — run the code-completion example with DeepSeek-Coder-V2-Lite-Base on a short Python snippet, then try the chat example with the Instruct variant.
Qwen3-Coder - Alibaba's open code model series for agentic coding, up to 1M tokens of context
Maid is a free and open source application for interfacing with llama.cpp models locally, and with Anthropic, DeepSeek, Ollama, Mistral and OpenAI models rem…
Run AI models locally on your machine with node.js bindings for llama.cpp. Enforce a JSON schema on the model output on the generation level
Private AI chat and model studio that runs in your browser via WebGPU.
vLLM-Omni — efficient inference for omni-modality models: image, audio, video, diffusion and policies.
GGUF Quantization support for native ComfyUI models
Другие проекты из категории «Локальные LLM» с наибольшим числом звёзд в каталоге LibreAI — реалистичный список альтернатив, если DeepSeek-Coder-V2 вам не подходит.
The simplest way to run open language models on your own machine.
User-friendly AI Interface (Supports Ollama, OpenAI API, ...)
LLM inference in C/C++
A high-throughput and memory-efficient inference and serving engine for LLMs
🤯 LobeHub is your Chief Agent Operator, organizing your agents into 7×24 operations by hiring, scheduling, and reporting on your entire AI team.
Соседние проекты из категории «Локальные LLM» в каталоге, ближайшие по числу звёзд к DeepSeek-Coder-V2 — продолжайте изучать полку по обе стороны от него.
vLLM-Omni — efficient inference for omni-modality models: image, audio, video, diffusion and policies.
Quantization, kernels, runtime and inference engine for mobiles, wearables, smart home and robots
Fine-tune LLMs from one YAML. Layer streaming trains an 8B model on a 4 GB laptop GPU.
⚡ Pure-Rust WebGPU inference engine — OpenAI-API compatible, GGUF native, runs on any GPU. No Python. No llama.cpp. Single binary.
KoboldCpp - Run GGUF AI models locally with a KoboldAI UI. One File. Zero Install
Discover and run local AI apps, serving optimized LLMs from your own GPU and NPU.
Бесплатно и открыто — забирайте исходный код, ставьте звезду репозиторию и начинайте создавать.