LLM local
ramalama
RamaLama is an open-source developer tool that simplifies the local serving of AI models from any source and facilitates their use for inference in productio…
★ 3,1k estrellasMIT
Ver la guía →// local-llm · openai/gpt-oss
GPT-OSS
GPT-OSS es un proyecto de LLM local de código abierto de openai (openai/gpt-oss en GitHub). Tiene 20.511 estrellas en GitHub y 2.190 bifurcaciones, se publica bajo la licencia Apache-2.0 y se actualizó por última vez en GitHub el 29 de octubre de 2026.
GPT-OSS — OpenAI open-weight language models: gpt-oss-120b for production reasoning and gpt-oss-20b for local use, Apache-2.0
★ 20,5k estrellas⑂ 2,2k bifurcacionesApache-2.0actualizado oct 2026
▦ visual: el README de este proyecto no incluye ninguna captura incrustable,
así que aquí va un marcador. La guía de abajo está completa de todos modos.
así que aquí va un marcador. La guía de abajo está completa de todos modos.
// datos del proyecto
Datos del proyecto
- Repositorio
- openai/gpt-oss
- Estrellas en GitHub
- 20.511
- Bifurcaciones
- 2.190
- Licencia
- Apache-2.0
- Última actualización en GitHub
- 29 de octubre de 2026
- Categoría
- LLM local
- Tipo de ficha
- Guía de instalación completa
- Sitio oficial
- OpenAI open models
// cómo funciona
Cómo funciona
- Pick the model that fits your hardware
Choose gpt-oss-120b for production-grade reasoning on a single 80GB GPU, or gpt-oss-20b for lower latency and local or specialised use within 16GB of memory.
- Install a serving route
Use the Transformers library, vLLM, or the reference gpt-oss package from the repository — all three routes are documented in the README.
- Use the harmony format
Send prompts in the harmony response format, via the Transformers chat template or the openai-harmony package; using the models without it will not work correctly.
// características
Por qué gusta tanto
Two genuinely open-weight OpenAI models
A 120b production model and a 20b local model, both Apache-2.0, with weights on Hugging Face.
MXFP4 quantisation built in
Post-training MXFP4 quantisation of the mixture-of-experts weights is what lets the 120b model run on one 80GB GPU and the 20b model in 16GB.
High-reasoning design
The 120b model is positioned for general-purpose, high-reasoning use cases; the 20b model trades size for latency and local deployment.
Standard serving routes
Transformers, vLLM and the reference package are all documented, so the models drop into existing inference stacks.
Advertisement
// ejecútalo
Ejecútalo tú mismo
- Install the reference package
The repository reference implementation from PyPI.
pip install gpt-oss - Run with Transformers
The README route: the Transformers pipeline with the harmony chat template applied automatically.
https://cookbook.openai.com/articles/gpt-oss/run-transformers - Serve with vLLM
The README serving route for the smaller model.
vllm serve openai/gpt-oss-20b
// casos de uso
Qué puedes crear
- Running an open-weight OpenAI model locally or on a single GPU server
- High-reasoning chat, coding and agent workloads with the 120b model
- Low-latency local or specialised deployments with the 20b model
- Studying open-weight mixture-of-experts models and MXFP4 quantisation
Advertisement
// preguntas frecuentes
Preguntas frecuentes
Who makes GPT-OSS?
OpenAI, as its open-weight model family (gpt-oss-120b and gpt-oss-20b).
Is GPT-OSS free?
Yes — Apache-2.0 licensed open weights. You still pay for your own hardware or hosting.
What hardware do I need?
gpt-oss-120b fits a single 80GB GPU (H100 or MI300X class); gpt-oss-20b runs within 16GB of memory.
What is the harmony format?
The response format both models were trained with. The README is explicit: use it (via the Transformers chat template or the openai-harmony package) or the models will not work correctly.
How do I get started?
Pick a model for your hardware, install via Transformers, vLLM or the gpt-oss package, and serve it with the harmony format.
// sigue explorando
Herramientas similares
LLM local
atomic-agent
Atomic Agent is a local-first AI agent. Runs open-weight models on your own machine via llama.cpp.
★ 2,7k estrellasMIT
Ver la guía →LLM local
OLMo-core
OLMo-core — the current PyTorch building blocks for the OLMo open language model ecosystem, Apache-2.0
★ 1,8k estrellasApache-2.0
Ver la guía →LLM local
WhisperJAV
Noise-robust subtitle generator powered by Whisper and local ASR models.
★ 2,3k estrellasMIT
Ver la guía →LLM local
FastChat
An open platform for training, serving, and evaluating large language models. Release repo for Vicuna and Chatbot Arena.
★ 39,6k estrellasApache-2.0
Ver la guía →LLM local
ollama-gui
Clean web interface for chatting with your local Ollama models.
★ 1,3k estrellasMIT
Ver la guía →// alternativas
Alternativas en LLM local
Los demás proyectos de LLM local con más estrellas del catálogo de LibreAI: la lista realista de alternativas si GPT-OSS no es lo que buscas.
- Ollama★ 182,1k
The simplest way to run open language models on your own machine.
- Open Webui★ 153,9k
User-friendly AI Interface (Supports Ollama, OpenAI API, ...)
- llama.cpp★ 130,2k
LLM inference in C/C++
- vLLM★ 93,1k
A high-throughput and memory-efficient inference and serving engine for LLMs
- lobehub★ 83,0k
🤯 LobeHub is your Chief Agent Operator, organizing your agents into 7×24 operations by hiring, scheduling, and reporting on your entire AI team.
// más en local-llm
Más en LLM local
Proyectos de LLM local cercanos en el catálogo, los más próximos en estrellas a GPT-OSS: sigue explorando a ambos lados.
- omlx★ 22,6k
A fast LLM inference server for Apple Silicon, managed from the macOS menu bar
- Web LLM★ 19,2k
High-performance In-browser LLM Inference Engine
- MaxKB★ 22,9k
🔥 MaxKB is an open-source platform for building enterprise-grade agents. 强大易用的开源企业级智能体平台。
- Qwen3-Coder★ 16,8k
Qwen3-Coder - Alibaba's open code model series for agentic coding, up to 1M tokens of context
- mlc-llm★ 23,2k
Compile and run LLMs anywhere - phones, laptops, browsers and GPUs - with one engine
- MNN★ 16,2k
Alibaba's lightweight, blazing-fast inference engine for on-device LLMs and edge AI.
Consigue GPT-OSS
Gratis y abierto: consigue el código fuente, dale una estrella al repositorio y empieza a crear.