[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"project-71970":3},{"id":4,"name":5,"fullName":6,"owner":7,"repo":5,"description":8,"homepage":9,"htmlUrl":10,"language":11,"languages":10,"totalLinesOfCode":10,"stars":12,"forks":13,"watchers":14,"openIssues":15,"contributorsCount":16,"subscribersCount":16,"size":16,"stars1d":17,"stars7d":18,"stars30d":19,"stars90d":16,"forks30d":16,"starsTrendScore":20,"compositeScore":21,"rankGlobal":10,"rankLanguage":10,"license":22,"archived":23,"fork":23,"defaultBranch":24,"hasWiki":23,"hasPages":23,"topics":25,"createdAt":10,"pushedAt":10,"updatedAt":38,"readmeContent":39,"aiSummary":40,"trendingCount":16,"starSnapshotCount":16,"syncStatus":41,"lastSyncTime":42,"discoverSource":43},71970,"oumi","oumi-ai\u002Foumi","oumi-ai","Easily fine-tune, evaluate and deploy Gemma 4, Qwen3.5, Qwen3.6, gpt-oss, DeepSeek-R1, or any open source LLM \u002F VLM!","https:\u002F\u002Foumi.ai",null,"Python",9316,778,65,11,0,7,68,90,21,39.67,"Apache License 2.0",false,"main",[26,27,28,29,30,31,32,33,34,35,36,37],"dpo","evaluation","fine-tuning","gpt-oss","gpt-oss-120b","gpt-oss-20b","inference","llama","llms","sft","slms","vlms","2026-06-12 02:02:56","![Oumi Logo](https:\u002F\u002Fgithub.com\u002Foumi-ai\u002Foumi\u002Fraw\u002Fmain\u002Fdocs\u002F_static\u002Flogo\u002Fheader_readme.svg)\n\n[![Documentation](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FDocumentation-oumi-blue.svg)](https:\u002F\u002Foumi.ai\u002Fdocs\u002Fen\u002Flatest\u002Findex.html)\n[![Blog](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FBlog-oumi-blue.svg)](https:\u002F\u002Foumi.ai\u002Fblog)\n[![Twitter](https:\u002F\u002Fimg.shields.io\u002Ftwitter\u002Ffollow\u002FOumi_PBC)](https:\u002F\u002Fx.com\u002FOumi_PBC)\n[![Discord](https:\u002F\u002Fimg.shields.io\u002Fdiscord\u002F1286348126797430814?label=Discord)](https:\u002F\u002Fdiscord.gg\u002Foumi)\n[![PyPI version](https:\u002F\u002Fbadge.fury.io\u002Fpy\u002Foumi.svg)](https:\u002F\u002Fbadge.fury.io\u002Fpy\u002Foumi)\n[![License](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FLicense-Apache%202.0-blue.svg)](https:\u002F\u002Fopensource.org\u002Flicenses\u002FApache-2.0)\n[![Tests](https:\u002F\u002Fgithub.com\u002Foumi-ai\u002Foumi\u002Factions\u002Fworkflows\u002Fpretest.yaml\u002Fbadge.svg?branch=main)](https:\u002F\u002Fgithub.com\u002Foumi-ai\u002Foumi\u002Factions\u002Fworkflows\u002Fpretest.yaml)\n[![GPU Tests](https:\u002F\u002Fgithub.com\u002Foumi-ai\u002Foumi\u002Factions\u002Fworkflows\u002Fgpu_tests.yaml\u002Fbadge.svg?branch=main)](https:\u002F\u002Fgithub.com\u002Foumi-ai\u002Foumi\u002Factions\u002Fworkflows\u002Fgpu_tests.yaml)\n[![GitHub Repo stars](https:\u002F\u002Fimg.shields.io\u002Fgithub\u002Fstars\u002Foumi-ai\u002Foumi)](https:\u002F\u002Fgithub.com\u002Foumi-ai\u002Foumi\u002Fstargazers)\n[![Code style: black](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002Fcode%20style-black-000000.svg)](https:\u002F\u002Fgithub.com\u002Fpsf\u002Fblack)\n[![pre-commit](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002Fpre--commit-enabled-brightgreen?logo=pre-commit)](https:\u002F\u002Fgithub.com\u002Fpre-commit\u002Fpre-commit)\n[![About](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FAbout-oumi-blue.svg)](https:\u002F\u002Foumi.ai)\n\n### Everything you need to build state-of-the-art foundation models, end-to-end\n\n\u003Cp align=\"center\">\n  \u003Ca href=\"https:\u002F\u002Ftrendshift.io\u002Frepositories\u002F12865\">\n    \u003Cimg alt=\"GitHub trending\" src=\"https:\u002F\u002Ftrendshift.io\u002Fapi\u002Fbadge\u002Frepositories\u002F12865\" \u002F>\n  \u003C\u002Fa>\n\u003C\u002Fp>\n\n## 🔥 News\n\n- [2026\u002F05] [Oumi v0.8 released](https:\u002F\u002Fgithub.com\u002Foumi-ai\u002Foumi\u002Freleases\u002Ftag\u002Fv0.8) with `oumi deploy` CLI for dedicated inference endpoints, an `oumi-mcp` MCP server for Claude\u002FCursor integration, batch API support across Anthropic\u002FFireworks\u002FTogether, and Transformers v5 \u002F TRL \u002F vLLM dependency upgrades\n- [2026\u002F03] Upgraded to Transformers v5, TRL v0.30, vLLM v0.19, and veRL v0.7 compatibility\n- [2026\u002F03] [MCP Integration Phase 1](https:\u002F\u002Fgithub.com\u002Foumi-ai\u002Foumi\u002Fpull\u002F2234): package scaffold and dependencies for MCP server support\n- [2026\u002F03] New: `oumi deploy` command for deploying oumi models dedicated inference endpoints on fireworks.ai and parasail\n- [2026\u002F03] Added support for Qwen3.5 model family\n- [2026\u002F03] Inference engines received multiple improvements: list_models api, improved error reporting\n- [2026\u002F02] [Preview of using the Oumi Platform and Lambda to fine-tune and deploy a 4B model for user intent classification](https:\u002F\u002Fyoutu.be\u002F0XpfYRpd_FA)\n- [2026\u002F02] [Lambda and Oumi partner for end-to-end custom model development](https:\u002F\u002Fblog.oumi.ai\u002Fp\u002Flambda-and-oumi-partner-for-end-to)\n- [2025\u002F12] [Oumi v0.6.0 released](https:\u002F\u002Fgithub.com\u002Foumi-ai\u002Foumi\u002Freleases\u002Ftag\u002Fv0.6.0) with Python 3.13 support, `oumi analyze` CLI command, TRL 0.26+ support, and more\n- [2025\u002F12] [WeMakeDevs AI Agents Assemble Hackathon: Oumi webinar on Finetuning for Text-to-SQL](https:\u002F\u002Fwww.youtube.com\u002Fwatch?v=6wPikqRZ7bQ&t=3203s)\n- [2025\u002F12] [Oumi co-sponsors WeMakeDevs AI Agents Assemble Hackathon with over 2000 project submissions](https:\u002F\u002Fwww.wemakedevs.org\u002Fhackathons\u002Fassemblehack25)\n- [2025\u002F11] [Oumi v0.5.0 released](https:\u002F\u002Fgithub.com\u002Foumi-ai\u002Foumi\u002Freleases\u002Ftag\u002Fv0.5) with advanced data synthesis, hyperparameter tuning automation, support for OpenEnv, and more\n\n\u003Cdetails>\n\u003Csummary>Older updates\u003C\u002Fsummary>\n\n- [2025\u002F11] [Example notebook to perform RLVF fine-tuning with OpenEnv](https:\u002F\u002Fgithub.com\u002Foumi-ai\u002Foumi\u002Fblob\u002Fmain\u002Fnotebooks\u002FOumi%20-%20OpenEnv%20GRPO%20with%20trl.ipynb), an open source library from the Meta PyTorch team for creating, deploying, and distributing agentic RL environments\n- [2025\u002F10] [Oumi v0.4.1](https:\u002F\u002Fgithub.com\u002Foumi-ai\u002Foumi\u002Freleases\u002Ftag\u002Fv0.4.1) and [v0.4.2](https:\u002F\u002Fgithub.com\u002Foumi-ai\u002Foumi\u002Freleases\u002Ftag\u002Fv0.4.2) released] with support for Qwen3-VL and Transformers v4.56, data synthesis documentation and examples, and many bug fixes\n- [2025\u002F09] [Oumi v0.4.0 released](https:\u002F\u002Fgithub.com\u002Foumi-ai\u002Foumi\u002Freleases\u002Ftag\u002Fv0.4.0) with DeepSpeed support, a Hugging Face Hub cache management tool, KTO\u002FVision DPO trainer support\n- [2025\u002F08] Training and inference support for OpenAI's `gpt-oss-20b` and `gpt-oss-120b`: [recipes here](https:\u002F\u002Fgithub.com\u002Foumi-ai\u002Foumi\u002Ftree\u002Fmain\u002Fconfigs\u002Frecipes\u002Fgpt_oss)\n- [2025\u002F08] Aug 14 Webinar - [OpenAI's gpt-oss: Separating the Substance from the Hype](https:\u002F\u002Fyoutu.be\u002Fg1PkAV7fXn0).\n- [2025\u002F08] [Oumi v0.3.0 released](https:\u002F\u002Fgithub.com\u002Foumi-ai\u002Foumi\u002Freleases\u002Ftag\u002Fv0.3.0) with model quantization (AWQ), an improved LLM-as-a-Judge API, and Adaptive Inference\n- [2025\u002F07] Recipe for [Qwen3 235B](https:\u002F\u002Fgithub.com\u002Foumi-ai\u002Foumi\u002Fblob\u002Fmain\u002Fconfigs\u002Frecipes\u002Fqwen3\u002Finference\u002F235b_a22b_together_infer.yaml)\n- [2025\u002F07] July 24 webinar: [\"Training a State-of-the-art Agent LLM with Oumi + Lambda\"](https:\u002F\u002Fyoutu.be\u002Ff3SU_heBP54)\n- [2025\u002F06] [Oumi v0.2.0 released](https:\u002F\u002Fgithub.com\u002Foumi-ai\u002Foumi\u002Freleases\u002Ftag\u002Fv0.2.0) with support for GRPO fine-tuning, a plethora of new model support, and much more\n- [2025\u002F06] Announcement of [Data Curation for Vision Language Models (DCVLR) competition](https:\u002F\u002Foumi.ai\u002Fblog\u002Fposts\u002Fannouncing-dcvlr) at NeurIPS2025\n- [2025\u002F06] Recipes for training, inference, and eval with the newly released [Falcon-H1](https:\u002F\u002Fgithub.com\u002Foumi-ai\u002Foumi\u002Ftree\u002Fmain\u002Fconfigs\u002Frecipes\u002Ffalcon_h1) and [Falcon-E](https:\u002F\u002Fgithub.com\u002Foumi-ai\u002Foumi\u002Ftree\u002Fmain\u002Fconfigs\u002Frecipes\u002Ffalcon_e) models\n- [2025\u002F05] Support and recipes for [InternVL3 1B](https:\u002F\u002Fgithub.com\u002Foumi-ai\u002Foumi\u002Ftree\u002Fmain\u002Fconfigs\u002Frecipes\u002Fvision\u002Finternvl3)\n- [2025\u002F04] Added support for training and inference with Llama 4 models: Scout (17B activated, 109B total) and Maverick (17B activated, 400B total) variants, including full fine-tuning, LoRA, and QLoRA configurations\n- [2025\u002F04] Recipes for [Qwen3 model family](https:\u002F\u002Fgithub.com\u002Foumi-ai\u002Foumi\u002Ftree\u002Fmain\u002Fconfigs\u002Frecipes\u002Fqwen3)\n- [2025\u002F04] Introducing HallOumi: a State-of-the-Art Claim-Verification Model [(technical overview)](https:\u002F\u002Foumi.ai\u002Fblog\u002Fposts\u002Fintroducing-halloumi)\n- [2025\u002F04] Oumi now supports two new Vision-Language models: [Phi4](https:\u002F\u002Fgithub.com\u002Foumi-ai\u002Foumi\u002Ftree\u002Fmain\u002Fconfigs\u002Frecipes\u002Fvision\u002Fphi4) and [Qwen 2.5](https:\u002F\u002Fgithub.com\u002Foumi-ai\u002Foumi\u002Ftree\u002Fmain\u002Fconfigs\u002Frecipes\u002Fvision\u002Fqwen2_5_vl_3b)\n\n\u003C\u002Fdetails>\n\n## 🔎 About\n\nOumi is a fully open-source platform that streamlines the entire lifecycle of foundation models - from data preparation and training to evaluation and deployment. Whether you're developing on a laptop, launching large scale experiments on a cluster, or deploying models in production, Oumi provides the tools and workflows you need.\n\nWith Oumi, you can:\n\n- 🚀 Train and fine-tune models from 10M to 405B parameters using state-of-the-art techniques (SFT, LoRA, QLoRA, GRPO, and more)\n- 🤖 Work with both text and multimodal models (Llama, DeepSeek, Qwen, Phi, and others)\n- 🔄 Synthesize and curate training data with LLM judges\n- ⚡️ Deploy models efficiently with popular inference engines (vLLM, SGLang)\n- 📊 Evaluate models comprehensively across standard benchmarks\n- 🌎 Run anywhere - from laptops to clusters to clouds (AWS, Azure, GCP, Lambda, and more)\n- 🔌 Integrate with both open models and commercial APIs (OpenAI, Anthropic, Vertex AI, Together, Parasail, ...)\n\nAll with one consistent API, production-grade reliability, and all the flexibility you need for research.\n\nLearn more at [oumi.ai](https:\u002F\u002Foumi.ai\u002Fdocs), or jump right in with the [quickstart guide](https:\u002F\u002Foumi.ai\u002Fdocs\u002Fen\u002Flatest\u002Fget_started\u002Fquickstart.html).\n\n## 🚀 Getting Started\n\n| **Notebook** | **Try in Colab** | **Goal** |\n|----------|--------------|-------------|\n| **🎯 Getting Started: A Tour** | \u003Ca target=\"_blank\" href=\"https:\u002F\u002Fcolab.research.google.com\u002Fgithub\u002Foumi-ai\u002Foumi\u002Fblob\u002Fmain\u002Fnotebooks\u002FOumi - A Tour.ipynb\">\u003Cimg src=\"https:\u002F\u002Fcolab.research.google.com\u002Fassets\u002Fcolab-badge.svg\" alt=\"Open In Colab\"\u002F>\u003C\u002Fa> | Quick tour of core features: training, evaluation, inference, and job management |\n| **🔧 Model Finetuning Guide** | \u003Ca target=\"_blank\" href=\"https:\u002F\u002Fcolab.research.google.com\u002Fgithub\u002Foumi-ai\u002Foumi\u002Fblob\u002Fmain\u002Fnotebooks\u002FOumi - Finetuning Tutorial.ipynb\">\u003Cimg src=\"https:\u002F\u002Fcolab.research.google.com\u002Fassets\u002Fcolab-badge.svg\" alt=\"Open In Colab\"\u002F>\u003C\u002Fa> | End-to-end guide to LoRA tuning with data prep, training, and evaluation |\n| **📚 Model Distillation** | \u003Ca target=\"_blank\" href=\"https:\u002F\u002Fcolab.research.google.com\u002Fgithub\u002Foumi-ai\u002Foumi\u002Fblob\u002Fmain\u002Fnotebooks\u002FOumi - Distill a Large Model.ipynb\">\u003Cimg src=\"https:\u002F\u002Fcolab.research.google.com\u002Fassets\u002Fcolab-badge.svg\" alt=\"Open In Colab\"\u002F>\u003C\u002Fa> | Guide to distilling large models into smaller, efficient ones |\n| **📋 Model Evaluation** | \u003Ca target=\"_blank\" href=\"https:\u002F\u002Fcolab.research.google.com\u002Fgithub\u002Foumi-ai\u002Foumi\u002Fblob\u002Fmain\u002Fnotebooks\u002FOumi - Evaluation with Oumi.ipynb\">\u003Cimg src=\"https:\u002F\u002Fcolab.research.google.com\u002Fassets\u002Fcolab-badge.svg\" alt=\"Open In Colab\"\u002F>\u003C\u002Fa> | Comprehensive model evaluation using Oumi's evaluation framework |\n| **☁️ Remote Training** | \u003Ca target=\"_blank\" href=\"https:\u002F\u002Fcolab.research.google.com\u002Fgithub\u002Foumi-ai\u002Foumi\u002Fblob\u002Fmain\u002Fnotebooks\u002FOumi - Running Jobs Remotely.ipynb\">\u003Cimg src=\"https:\u002F\u002Fcolab.research.google.com\u002Fassets\u002Fcolab-badge.svg\" alt=\"Open In Colab\"\u002F>\u003C\u002Fa> | Launch and monitor training jobs on cloud (AWS, Azure, GCP, Lambda, etc.) platforms |\n| **📈 LLM-as-a-Judge** | \u003Ca target=\"_blank\" href=\"https:\u002F\u002Fcolab.research.google.com\u002Fgithub\u002Foumi-ai\u002Foumi\u002Fblob\u002Fmain\u002Fnotebooks\u002FOumi - Simple Judge.ipynb\">\u003Cimg src=\"https:\u002F\u002Fcolab.research.google.com\u002Fassets\u002Fcolab-badge.svg\" alt=\"Open In Colab\"\u002F>\u003C\u002Fa> | Filter and curate training data with built-in judges |\n\n## 🔧 Usage\n\n### Installation\n\nChoose the installation method that works best for you:\n\n\u003Cdetails open>\n\u003Csummary>\u003Cb>Using pip (Recommended)\u003C\u002Fb>\u003C\u002Fsummary>\n\n```bash\n# Basic installation\nuv pip install oumi\n\n# With GPU support\nuv pip install 'oumi[gpu]'\n\n# Latest development version\nuv pip install git+https:\u002F\u002Fgithub.com\u002Foumi-ai\u002Foumi.git\n```\n\nDon't have uv? [Install it](https:\u002F\u002Fdocs.astral.sh\u002Fuv\u002Fgetting-started\u002Finstallation\u002F) or use `pip` instead.\n\n\u003C\u002Fdetails>\n\n\u003Cdetails>\n\u003Csummary>\u003Cb>Using Docker\u003C\u002Fb>\u003C\u002Fsummary>\n\n```bash\n# Pull the latest image\ndocker pull ghcr.io\u002Foumi-ai\u002Foumi:latest\n\n# Run oumi commands\ndocker run --gpus all -it ghcr.io\u002Foumi-ai\u002Foumi:latest oumi --help\n\n# Train with a mounted config\ndocker run --gpus all -v $(pwd):\u002Fworkspace -it ghcr.io\u002Foumi-ai\u002Foumi:latest \\\n    oumi train --config \u002Fworkspace\u002Fmy_config.yaml\n```\n\n\u003C\u002Fdetails>\n\n\u003Cdetails>\n\u003Csummary>\u003Cb>Quick Install Script (Experimental)\u003C\u002Fb>\u003C\u002Fsummary>\n\nTry Oumi without setting up a Python environment. This installs Oumi in an isolated environment:\n\n```bash\ncurl -LsSf https:\u002F\u002Foumi.ai\u002Finstall.sh | bash\n```\n\n\u003C\u002Fdetails>\n\nFor more advanced installation options, see the [installation guide](https:\u002F\u002Foumi.ai\u002Fdocs\u002Fen\u002Flatest\u002Fget_started\u002Finstallation.html).\n\n### Oumi CLI\n\nYou can quickly use the `oumi` command to train, evaluate, and infer models using one of the existing [recipes](\u002Fconfigs\u002Frecipes):\n\n```shell\n# Training\noumi train -c configs\u002Frecipes\u002Fsmollm\u002Fsft\u002F135m\u002Fquickstart_train.yaml\n\n# Evaluation\noumi evaluate -c configs\u002Frecipes\u002Fsmollm\u002Fevaluation\u002F135m\u002Fquickstart_eval.yaml\n\n# Inference\noumi infer -c configs\u002Frecipes\u002Fsmollm\u002Finference\u002F135m_infer.yaml --interactive\n```\n\nFor more advanced options, see the [training](https:\u002F\u002Foumi.ai\u002Fdocs\u002Fen\u002Flatest\u002Fuser_guides\u002Ftrain\u002Ftrain.html), [evaluation](https:\u002F\u002Foumi.ai\u002Fdocs\u002Fen\u002Flatest\u002Fuser_guides\u002Fevaluate\u002Fevaluate.html), [inference](https:\u002F\u002Foumi.ai\u002Fdocs\u002Fen\u002Flatest\u002Fuser_guides\u002Finfer\u002Finfer.html), and [llm-as-a-judge](https:\u002F\u002Foumi.ai\u002Fdocs\u002Fen\u002Flatest\u002Fuser_guides\u002Fjudge\u002Fjudge.html) guides.\n\n### Running Jobs Remotely\n\nYou can run jobs remotely on cloud platforms (AWS, Azure, GCP, Lambda, etc.) using the `oumi launch` command:\n\n```shell\n# GCP\noumi launch up -c configs\u002Frecipes\u002Fsmollm\u002Fsft\u002F135m\u002Fquickstart_gcp_job.yaml\n\n# AWS\noumi launch up -c configs\u002Frecipes\u002Fsmollm\u002Fsft\u002F135m\u002Fquickstart_gcp_job.yaml --resources.cloud aws\n\n# Azure\noumi launch up -c configs\u002Frecipes\u002Fsmollm\u002Fsft\u002F135m\u002Fquickstart_gcp_job.yaml --resources.cloud azure\n\n# Lambda\noumi launch up -c configs\u002Frecipes\u002Fsmollm\u002Fsft\u002F135m\u002Fquickstart_gcp_job.yaml --resources.cloud lambda\n```\n\n**Note:** Oumi is in \u003Cins>beta\u003C\u002Fins> and under active development. The core features are stable, but some advanced features might change as the platform improves.\n\n## 💻 Why use Oumi?\n\nIf you need a comprehensive platform for training, evaluating, or deploying models, Oumi is a great choice.\n\nHere are some of the key features that make Oumi stand out:\n\n- 🔧 **Zero Boilerplate**: Get started in minutes with ready-to-use recipes for popular models and workflows. No need to write training loops or data pipelines.\n- 🏢 **Enterprise-Grade**: Built and validated by teams training models at scale\n- 🎯 **Research Ready**: Perfect for ML research with easily reproducible experiments, and flexible interfaces for customizing each component.\n- 🌐 **Broad Model Support**: Works with most popular model architectures - from tiny models to the largest ones, text-only to multimodal.\n- 🚀 **SOTA Performance**: Native support for distributed training techniques (FSDP, DeepSpeed, DDP) and optimized inference engines (vLLM, SGLang).\n- 🤝 **Community First**: 100% open source with an active community. No vendor lock-in, no strings attached.\n\n## 📚 Examples &  Recipes\n\nExplore the growing collection of ready-to-use configurations for state-of-the-art models and training workflows:\n\n**Note:** These configurations are not an exhaustive list of what's supported, simply examples to get you started. You can find a more exhaustive list of supported [models](https:\u002F\u002Foumi.ai\u002Fdocs\u002Fen\u002Flatest\u002Fresources\u002Fmodels\u002Fsupported_models.html), and datasets ([supervised fine-tuning](https:\u002F\u002Foumi.ai\u002Fdocs\u002Fen\u002Flatest\u002Fresources\u002Fdatasets\u002Fsft_datasets.html), [pre-training](https:\u002F\u002Foumi.ai\u002Fdocs\u002Fen\u002Flatest\u002Fresources\u002Fdatasets\u002Fpretraining_datasets.html), [preference tuning](https:\u002F\u002Foumi.ai\u002Fdocs\u002Fen\u002Flatest\u002Fresources\u002Fdatasets\u002Fpreference_datasets.html), and [vision-language finetuning](https:\u002F\u002Foumi.ai\u002Fdocs\u002Fen\u002Flatest\u002Fresources\u002Fdatasets\u002Fvl_sft_datasets.html)) in the oumi documentation.\n\n### Qwen Family\n\n| Model | Example Configurations |\n|-------|------------------------|\n| Qwen3-Next 80B A3B | [LoRA](\u002Fconfigs\u002Frecipes\u002Fqwen3_next\u002Fsft\u002F80b_a3b_lora\u002Ftrain.yaml) • [Inference](\u002Fconfigs\u002Frecipes\u002Fqwen3_next\u002Finference\u002F80b_a3b_infer.yaml) • [Inference (Instruct)](\u002Fconfigs\u002Frecipes\u002Fqwen3_next\u002Finference\u002F80b_a3b_instruct_infer.yaml) • [Evaluation](\u002Fconfigs\u002Frecipes\u002Fqwen3_next\u002Fevaluation\u002F80b_a3b_eval.yaml) |\n| Qwen3 30B A3B | [LoRA](\u002Fconfigs\u002Frecipes\u002Fqwen3\u002Fsft\u002F30b_a3b_lora\u002Ftrain.yaml) • [Inference](\u002Fconfigs\u002Frecipes\u002Fqwen3\u002Finference\u002F30b_a3b_infer.yaml) • [Evaluation](\u002Fconfigs\u002Frecipes\u002Fqwen3\u002Fevaluation\u002F30b_a3b_eval.yaml) |\n| Qwen3 32B | [LoRA](\u002Fconfigs\u002Frecipes\u002Fqwen3\u002Fsft\u002F32b_lora\u002Ftrain.yaml) • [Inference](\u002Fconfigs\u002Frecipes\u002Fqwen3\u002Finference\u002F32b_infer.yaml) • [Evaluation](\u002Fconfigs\u002Frecipes\u002Fqwen3\u002Fevaluation\u002F32b_eval.yaml) |\n| Qwen3 14B | [LoRA](\u002Fconfigs\u002Frecipes\u002Fqwen3\u002Fsft\u002F14b_lora\u002Ftrain.yaml) • [Inference](\u002Fconfigs\u002Frecipes\u002Fqwen3\u002Finference\u002F14b_infer.yaml) • [Evaluation](\u002Fconfigs\u002Frecipes\u002Fqwen3\u002Fevaluation\u002F14b_eval.yaml) |\n| Qwen3 8B | [FFT](\u002Fconfigs\u002Frecipes\u002Fqwen3\u002Fsft\u002F8b_full\u002Ftrain.yaml) • [Inference](\u002Fconfigs\u002Frecipes\u002Fqwen3\u002Finference\u002F8b_infer.yaml) • [Evaluation](\u002Fconfigs\u002Frecipes\u002Fqwen3\u002Fevaluation\u002F8b_eval.yaml) |\n| Qwen3 4B | [FFT](\u002Fconfigs\u002Frecipes\u002Fqwen3\u002Fsft\u002F4b_full\u002Ftrain.yaml) • [Inference](\u002Fconfigs\u002Frecipes\u002Fqwen3\u002Finference\u002F4b_infer.yaml) • [Evaluation](\u002Fconfigs\u002Frecipes\u002Fqwen3\u002Fevaluation\u002F4b_eval.yaml) |\n| Qwen3 1.7B | [FFT](\u002Fconfigs\u002Frecipes\u002Fqwen3\u002Fsft\u002F1.7b_full\u002Ftrain.yaml) • [Inference](\u002Fconfigs\u002Frecipes\u002Fqwen3\u002Finference\u002F1.7b_infer.yaml) • [Evaluation](\u002Fconfigs\u002Frecipes\u002Fqwen3\u002Fevaluation\u002F1.7b_eval.yaml) |\n| Qwen3 0.6B | [FFT](\u002Fconfigs\u002Frecipes\u002Fqwen3\u002Fsft\u002F0.6b_full\u002Ftrain.yaml) • [Inference](\u002Fconfigs\u002Frecipes\u002Fqwen3\u002Finference\u002F0.6b_infer.yaml) • [Evaluation](\u002Fconfigs\u002Frecipes\u002Fqwen3\u002Fevaluation\u002F0.6b_eval.yaml) |\n| QwQ 32B | [FFT](\u002Fconfigs\u002Frecipes\u002Fqwq\u002Fsft\u002Ffull_train.yaml) • [LoRA](\u002Fconfigs\u002Frecipes\u002Fqwq\u002Fsft\u002Flora_train.yaml) • [QLoRA](\u002Fconfigs\u002Frecipes\u002Fqwq\u002Fsft\u002Fqlora_train.yaml) • [Inference](\u002Fconfigs\u002Frecipes\u002Fqwq\u002Finference\u002Finfer.yaml) • [Evaluation](\u002Fconfigs\u002Frecipes\u002Fqwq\u002Fevaluation\u002Feval.yaml) |\n| Qwen2.5-VL 3B | [SFT](\u002Fconfigs\u002Frecipes\u002Fvision\u002Fqwen2_5_vl_3b\u002Fsft\u002Ffull\u002Ftrain.yaml) • [LoRA](\u002Fconfigs\u002Frecipes\u002Fvision\u002Fqwen2_5_vl_3b\u002Fsft\u002Flora\u002Ftrain.yaml)• [Inference (vLLM)](configs\u002Frecipes\u002Fvision\u002Fqwen2_5_vl_3b\u002Finference\u002Fvllm_infer.yaml) • [Inference](configs\u002Frecipes\u002Fvision\u002Fqwen2_5_vl_3b\u002Finference\u002Finfer.yaml) |\n| Qwen2-VL 2B | [SFT](\u002Fconfigs\u002Frecipes\u002Fvision\u002Fqwen2_vl_2b\u002Fsft\u002Ffull\u002Ftrain.yaml) • [LoRA](\u002Fconfigs\u002Frecipes\u002Fvision\u002Fqwen2_vl_2b\u002Fsft\u002Flora\u002Ftrain.yaml) • [Inference (vLLM)](configs\u002Frecipes\u002Fvision\u002Fqwen2_vl_2b\u002Finference\u002Fvllm_infer.yaml) • [Inference (SGLang)](configs\u002Frecipes\u002Fvision\u002Fqwen2_vl_2b\u002Finference\u002Fsglang_infer.yaml) • [Inference](configs\u002Frecipes\u002Fvision\u002Fqwen2_vl_2b\u002Finference\u002Finfer.yaml) • [Evaluation](configs\u002Frecipes\u002Fvision\u002Fqwen2_vl_2b\u002Fevaluation\u002Feval.yaml) |\n\n### 🐋 DeepSeek R1 Family\n\n| Model | Example Configurations |\n|-------|------------------------|\n| DeepSeek R1 671B | [Inference (Together AI)](configs\u002Frecipes\u002Fdeepseek_r1\u002Finference\u002F671b_together_infer.yaml) |\n| Distilled Llama 8B | [FFT](\u002Fconfigs\u002Frecipes\u002Fdeepseek_r1\u002Fsft\u002Fdistill_llama_8b\u002Ffull_train.yaml) • [LoRA](\u002Fconfigs\u002Frecipes\u002Fdeepseek_r1\u002Fsft\u002Fdistill_llama_8b\u002Flora_train.yaml) • [QLoRA](\u002Fconfigs\u002Frecipes\u002Fdeepseek_r1\u002Fsft\u002Fdistill_llama_8b\u002Fqlora_train.yaml) • [Inference](configs\u002Frecipes\u002Fdeepseek_r1\u002Finference\u002Fdistill_llama_8b_infer.yaml) • [Evaluation](\u002Fconfigs\u002Frecipes\u002Fdeepseek_r1\u002Fevaluation\u002Fdistill_llama_8b\u002Feval.yaml) |\n| Distilled Llama 70B | [FFT](\u002Fconfigs\u002Frecipes\u002Fdeepseek_r1\u002Fsft\u002Fdistill_llama_70b\u002Ffull_train.yaml) • [LoRA](\u002Fconfigs\u002Frecipes\u002Fdeepseek_r1\u002Fsft\u002Fdistill_llama_70b\u002Flora_train.yaml) • [QLoRA](\u002Fconfigs\u002Frecipes\u002Fdeepseek_r1\u002Fsft\u002Fdistill_llama_70b\u002Fqlora_train.yaml) • [Inference](configs\u002Frecipes\u002Fdeepseek_r1\u002Finference\u002Fdistill_llama_70b_infer.yaml) • [Evaluation](\u002Fconfigs\u002Frecipes\u002Fdeepseek_r1\u002Fevaluation\u002Fdistill_llama_70b\u002Feval.yaml) |\n| Distilled Qwen 1.5B | [FFT](\u002Fconfigs\u002Frecipes\u002Fdeepseek_r1\u002Fsft\u002Fdistill_qwen_1_5b\u002Ffull_train.yaml) • [LoRA](\u002Fconfigs\u002Frecipes\u002Fdeepseek_r1\u002Fsft\u002Fdistill_qwen_1_5b\u002Flora_train.yaml) • [Inference](configs\u002Frecipes\u002Fdeepseek_r1\u002Finference\u002Fdistill_qwen_1_5b_infer.yaml) • [Evaluation](\u002Fconfigs\u002Frecipes\u002Fdeepseek_r1\u002Fevaluation\u002Fdistill_qwen_1_5b\u002Feval.yaml) |\n| Distilled Qwen 32B | [LoRA](\u002Fconfigs\u002Frecipes\u002Fdeepseek_r1\u002Fsft\u002Fdistill_qwen_32b\u002Flora_train.yaml) • [Inference](configs\u002Frecipes\u002Fdeepseek_r1\u002Finference\u002Fdistill_qwen_32b_infer.yaml) • [Evaluation](\u002Fconfigs\u002Frecipes\u002Fdeepseek_r1\u002Fevaluation\u002Fdistill_qwen_32b\u002Feval.yaml) |\n\n### 🦙 Llama Family\n\n| Model | Example Configurations |\n|-------|------------------------|\n| Llama 4 Scout Instruct 17B | [FFT](\u002Fconfigs\u002Frecipes\u002Fllama4\u002Fsft\u002Fscout_instruct_full\u002Ftrain.yaml) • [LoRA](\u002Fconfigs\u002Frecipes\u002Fllama4\u002Fsft\u002Fscout_instruct_lora\u002Ftrain.yaml) • [QLoRA](\u002Fconfigs\u002Frecipes\u002Fllama4\u002Fsft\u002Fscout_instruct_qlora\u002Ftrain.yaml) • [Inference (vLLM)](\u002Fconfigs\u002Frecipes\u002Fllama4\u002Finference\u002Fscout_instruct_vllm_infer.yaml) • [Inference](\u002Fconfigs\u002Frecipes\u002Fllama4\u002Finference\u002Fscout_instruct_infer.yaml) • [Inference (Together.ai)](\u002Fconfigs\u002Frecipes\u002Fllama4\u002Finference\u002Fscout_instruct_together_infer.yaml) |\n| Llama 4 Scout 17B | [FFT](\u002Fconfigs\u002Frecipes\u002Fllama4\u002Fsft\u002Fscout_base_full\u002Ftrain.yaml)  |\n| Llama 3.1 8B | [FFT](\u002Fconfigs\u002Frecipes\u002Fllama3_1\u002Fsft\u002F8b_full\u002Ftrain.yaml) • [LoRA](\u002Fconfigs\u002Frecipes\u002Fllama3_1\u002Fsft\u002F8b_lora\u002Ftrain.yaml) • [QLoRA](\u002Fconfigs\u002Frecipes\u002Fllama3_1\u002Fsft\u002F8b_qlora\u002Ftrain.yaml) • [Pre-training](\u002Fconfigs\u002Frecipes\u002Fllama3_1\u002Fpretraining\u002F8b\u002Ftrain.yaml) • [Inference (vLLM)](configs\u002Frecipes\u002Fllama3_1\u002Finference\u002F8b_rvllm_infer.yaml) • [Inference](\u002Fconfigs\u002Frecipes\u002Fllama3_1\u002Finference\u002F8b_infer.yaml) • [Evaluation](\u002Fconfigs\u002Frecipes\u002Fllama3_1\u002Fevaluation\u002F8b_eval.yaml) |\n| Llama 3.1 70B | [FFT](\u002Fconfigs\u002Frecipes\u002Fllama3_1\u002Fsft\u002F70b_full\u002Ftrain.yaml) • [LoRA](\u002Fconfigs\u002Frecipes\u002Fllama3_1\u002Fsft\u002F70b_lora\u002Ftrain.yaml) • [QLoRA](\u002Fconfigs\u002Frecipes\u002Fllama3_1\u002Fsft\u002F70b_qlora\u002Ftrain.yaml) • [Inference](\u002Fconfigs\u002Frecipes\u002Fllama3_1\u002Finference\u002F70b_infer.yaml) • [Evaluation](\u002Fconfigs\u002Frecipes\u002Fllama3_1\u002Fevaluation\u002F70b_eval.yaml) |\n| Llama 3.1 405B | [FFT](\u002Fconfigs\u002Frecipes\u002Fllama3_1\u002Fsft\u002F405b_full\u002Ftrain.yaml) • [LoRA](\u002Fconfigs\u002Frecipes\u002Fllama3_1\u002Fsft\u002F405b_lora\u002Ftrain.yaml) • [QLoRA](\u002Fconfigs\u002Frecipes\u002Fllama3_1\u002Fsft\u002F405b_qlora\u002Ftrain.yaml) |\n| Llama 3.2 1B | [FFT](\u002Fconfigs\u002Frecipes\u002Fllama3_2\u002Fsft\u002F1b_full\u002Ftrain.yaml) • [LoRA](\u002Fconfigs\u002Frecipes\u002Fllama3_2\u002Fsft\u002F1b_lora\u002Ftrain.yaml) • [QLoRA](\u002Fconfigs\u002Frecipes\u002Fllama3_2\u002Fsft\u002F1b_qlora\u002Ftrain.yaml) • [Inference (vLLM)](\u002Fconfigs\u002Frecipes\u002Fllama3_2\u002Finference\u002F1b_vllm_infer.yaml) • [Inference (SGLang)](\u002Fconfigs\u002Frecipes\u002Fllama3_2\u002Finference\u002F1b_sglang_infer.yaml) • [Inference](\u002Fconfigs\u002Frecipes\u002Fllama3_2\u002Finference\u002F1b_infer.yaml) • [Evaluation](\u002Fconfigs\u002Frecipes\u002Fllama3_2\u002Fevaluation\u002F1b_eval.yaml) |\n| Llama 3.2 3B | [FFT](\u002Fconfigs\u002Frecipes\u002Fllama3_2\u002Fsft\u002F3b_full\u002Ftrain.yaml) • [LoRA](\u002Fconfigs\u002Frecipes\u002Fllama3_2\u002Fsft\u002F3b_lora\u002Ftrain.yaml) • [QLoRA](\u002Fconfigs\u002Frecipes\u002Fllama3_2\u002Fsft\u002F3b_qlora\u002Ftrain.yaml) • [Inference (vLLM)](\u002Fconfigs\u002Frecipes\u002Fllama3_2\u002Finference\u002F3b_vllm_infer.yaml) • [Inference (SGLang)](\u002Fconfigs\u002Frecipes\u002Fllama3_2\u002Finference\u002F3b_sglang_infer.yaml) • [Inference](\u002Fconfigs\u002Frecipes\u002Fllama3_2\u002Finference\u002F3b_infer.yaml) • [Evaluation](\u002Fconfigs\u002Frecipes\u002Fllama3_2\u002Fevaluation\u002F3b_eval.yaml) |\n| Llama 3.3 70B | [FFT](\u002Fconfigs\u002Frecipes\u002Fllama3_3\u002Fsft\u002F70b_full\u002Ftrain.yaml) • [LoRA](\u002Fconfigs\u002Frecipes\u002Fllama3_3\u002Fsft\u002F70b_lora\u002Ftrain.yaml) • [QLoRA](\u002Fconfigs\u002Frecipes\u002Fllama3_3\u002Fsft\u002F70b_qlora\u002Ftrain.yaml) • [Inference (vLLM)](\u002Fconfigs\u002Frecipes\u002Fllama3_3\u002Finference\u002F70b_vllm_infer.yaml) • [Inference](\u002Fconfigs\u002Frecipes\u002Fllama3_3\u002Finference\u002F70b_infer.yaml) • [Evaluation](\u002Fconfigs\u002Frecipes\u002Fllama3_3\u002Fevaluation\u002F70b_eval.yaml) |\n| Llama 3.2 Vision 11B | [SFT](\u002Fconfigs\u002Frecipes\u002Fvision\u002Fllama3_2_vision\u002Fsft\u002F11b_full\u002Ftrain.yaml) • [Inference (vLLM)](\u002Fconfigs\u002Frecipes\u002Fvision\u002Fllama3_2_vision\u002Finference\u002F11b_rvllm_infer.yaml) • [Inference (SGLang)](\u002Fconfigs\u002Frecipes\u002Fvision\u002Fllama3_2_vision\u002Finference\u002F11b_sglang_infer.yaml) • [Evaluation](\u002Fconfigs\u002Frecipes\u002Fvision\u002Fllama3_2_vision\u002Fevaluation\u002F11b_eval.yaml) |\n\n### 🦅 Falcon family\n\n| Model | Example Configurations |\n|-------|------------------------|\n| [Falcon-H1](https:\u002F\u002Fhuggingface.co\u002Fcollections\u002Ftiiuae\u002Ffalcon-h1-6819f2795bc406da60fab8df) | [FFT](\u002Fconfigs\u002Frecipes\u002Ffalcon_h1\u002Fsft\u002F) • [Inference](\u002Fconfigs\u002Frecipes\u002Ffalcon_h1\u002Finference\u002F) • [Evaluation](\u002Fconfigs\u002Frecipes\u002Ffalcon_h1\u002Fevaluation\u002F) |\n| [Falcon-E (BitNet)](https:\u002F\u002Fhuggingface.co\u002Fcollections\u002Ftiiuae\u002Ffalcon-edge-series-6804fd13344d6d8a8fa71130) | [FFT](\u002Fconfigs\u002Frecipes\u002Ffalcon_e\u002Fsft\u002F) • [DPO](\u002Fconfigs\u002Frecipes\u002Ffalcon_e\u002Fdpo\u002F) • [Evaluation](\u002Fconfigs\u002Frecipes\u002Ffalcon_e\u002Fevaluation\u002F) |\n\n### 💎 Gemma 3 Family\n\n| Model | Example Configurations |\n|-------|------------------------|\n| Gemma 3 4B Instruct | [FFT](\u002Fconfigs\u002Frecipes\u002Fgemma3\u002Fsft\u002F4b_full\u002Ftrain.yaml) • [Inference](\u002Fconfigs\u002Frecipes\u002Fgemma3\u002Finference\u002F4b_instruct_infer.yaml) • [Evaluation](\u002Fconfigs\u002Frecipes\u002Fgemma3\u002Fevaluation\u002F4b\u002Feval.yaml) |\n| Gemma 3 12B Instruct | [LoRA](\u002Fconfigs\u002Frecipes\u002Fgemma3\u002Fsft\u002F12b_lora\u002Ftrain.yaml) • [Inference](\u002Fconfigs\u002Frecipes\u002Fgemma3\u002Finference\u002F12b_instruct_infer.yaml) • [Evaluation](\u002Fconfigs\u002Frecipes\u002Fgemma3\u002Fevaluation\u002F12b\u002Feval.yaml) |\n| Gemma 3 27B Instruct | [LoRA](\u002Fconfigs\u002Frecipes\u002Fgemma3\u002Fsft\u002F27b_lora\u002Ftrain.yaml) • [Inference](\u002Fconfigs\u002Frecipes\u002Fgemma3\u002Finference\u002F27b_instruct_infer.yaml) • [Evaluation](\u002Fconfigs\u002Frecipes\u002Fgemma3\u002Fevaluation\u002F27b\u002Feval.yaml) |\n\n### 🦉 OLMo 3 Family\n\n| Model | Example Configurations |\n|-------|------------------------|\n| OLMo 3 7B Instruct | [FFT](\u002Fconfigs\u002Frecipes\u002Folmo3\u002Fsft\u002F7b_full\u002Ftrain.yaml) • [Inference](\u002Fconfigs\u002Frecipes\u002Folmo3\u002Finference\u002F7b_infer.yaml) • [Evaluation](\u002Fconfigs\u002Frecipes\u002Folmo3\u002Fevaluation\u002F7b\u002Feval.yaml) |\n| OLMo 3 32B Instruct | [LoRA](\u002Fconfigs\u002Frecipes\u002Folmo3\u002Fsft\u002F32b_lora\u002Ftrain.yaml) • [Inference](\u002Fconfigs\u002Frecipes\u002Folmo3\u002Finference\u002F32b_infer.yaml) • [Evaluation](\u002Fconfigs\u002Frecipes\u002Folmo3\u002Fevaluation\u002F32b\u002Feval.yaml) |\n\n### 🎨 Vision Models\n\n| Model | Example Configurations |\n|-------|------------------------|\n| Llama 3.2 Vision 11B | [SFT](\u002Fconfigs\u002Frecipes\u002Fvision\u002Fllama3_2_vision\u002Fsft\u002F11b_full\u002Ftrain.yaml) • [LoRA](\u002Fconfigs\u002Frecipes\u002Fvision\u002Fllama3_2_vision\u002Fsft\u002F11b_lora\u002Ftrain.yaml) • [Inference (vLLM)](\u002Fconfigs\u002Frecipes\u002Fvision\u002Fllama3_2_vision\u002Finference\u002F11b_rvllm_infer.yaml) • [Inference (SGLang)](\u002Fconfigs\u002Frecipes\u002Fvision\u002Fllama3_2_vision\u002Finference\u002F11b_sglang_infer.yaml) • [Evaluation](\u002Fconfigs\u002Frecipes\u002Fvision\u002Fllama3_2_vision\u002Fevaluation\u002F11b_eval.yaml) |\n| LLaVA 7B | [SFT](\u002Fconfigs\u002Frecipes\u002Fvision\u002Fllava_7b\u002Fsft\u002Ftrain.yaml) • [Inference (vLLM)](configs\u002Frecipes\u002Fvision\u002Fllava_7b\u002Finference\u002Fvllm_infer.yaml) • [Inference](\u002Fconfigs\u002Frecipes\u002Fvision\u002Fllava_7b\u002Finference\u002Finfer.yaml) |\n| Phi3 Vision 4.2B | [SFT](\u002Fconfigs\u002Frecipes\u002Fvision\u002Fphi3\u002Fsft\u002Ffull\u002Ftrain.yaml) • [LoRA](\u002Fconfigs\u002Frecipes\u002Fvision\u002Fphi3\u002Fsft\u002Flora\u002Ftrain.yaml) • [Inference (vLLM)](configs\u002Frecipes\u002Fvision\u002Fphi3\u002Finference\u002Fvllm_infer.yaml) |\n| Phi4 Vision 5.6B | [SFT](\u002Fconfigs\u002Frecipes\u002Fvision\u002Fphi4\u002Fsft\u002Ffull\u002Ftrain.yaml) • [LoRA](\u002Fconfigs\u002Frecipes\u002Fvision\u002Fphi4\u002Fsft\u002Flora\u002Ftrain.yaml) • [Inference (vLLM)](configs\u002Frecipes\u002Fvision\u002Fphi4\u002Finference\u002Fvllm_infer.yaml) • [Inference](\u002Fconfigs\u002Frecipes\u002Fvision\u002Fphi4\u002Finference\u002Finfer.yaml) |\n| Qwen2-VL 2B | [SFT](\u002Fconfigs\u002Frecipes\u002Fvision\u002Fqwen2_vl_2b\u002Fsft\u002Ffull\u002Ftrain.yaml) • [LoRA](\u002Fconfigs\u002Frecipes\u002Fvision\u002Fqwen2_vl_2b\u002Fsft\u002Flora\u002Ftrain.yaml) • [Inference (vLLM)](configs\u002Frecipes\u002Fvision\u002Fqwen2_vl_2b\u002Finference\u002Fvllm_infer.yaml) • [Inference (SGLang)](configs\u002Frecipes\u002Fvision\u002Fqwen2_vl_2b\u002Finference\u002Fsglang_infer.yaml) • [Inference](configs\u002Frecipes\u002Fvision\u002Fqwen2_vl_2b\u002Finference\u002Finfer.yaml) • [Evaluation](configs\u002Frecipes\u002Fvision\u002Fqwen2_vl_2b\u002Fevaluation\u002Feval.yaml) |\n| Qwen3-VL 2B | [Inference](\u002Fconfigs\u002Frecipes\u002Fvision\u002Fqwen3_vl_2b\u002Finference\u002Finfer.yaml) |\n| Qwen3-VL 4B | [Inference](\u002Fconfigs\u002Frecipes\u002Fvision\u002Fqwen3_vl_4b\u002Finference\u002Finfer.yaml) |\n| Qwen3-VL 8B | [Inference](\u002Fconfigs\u002Frecipes\u002Fvision\u002Fqwen3_vl_8b\u002Finference\u002Finfer.yaml) |\n| Qwen2.5-VL 3B | [SFT](\u002Fconfigs\u002Frecipes\u002Fvision\u002Fqwen2_5_vl_3b\u002Fsft\u002Ffull\u002Ftrain.yaml) • [LoRA](\u002Fconfigs\u002Frecipes\u002Fvision\u002Fqwen2_5_vl_3b\u002Fsft\u002Flora\u002Ftrain.yaml)• [Inference (vLLM)](configs\u002Frecipes\u002Fvision\u002Fqwen2_5_vl_3b\u002Finference\u002Fvllm_infer.yaml) • [Inference](configs\u002Frecipes\u002Fvision\u002Fqwen2_5_vl_3b\u002Finference\u002Finfer.yaml) |\n| SmolVLM-Instruct 2B | [SFT](\u002Fconfigs\u002Frecipes\u002Fvision\u002Fsmolvlm\u002Fsft\u002Ffull\u002Ftrain.yaml) • [LoRA](\u002Fconfigs\u002Frecipes\u002Fvision\u002Fsmolvlm\u002Fsft\u002Flora\u002Ftrain.yaml) |\n\n### 🔍 Even more options\n\nThis section lists all the language models that can be used with Oumi. Thanks to the integration with the [🤗 Transformers](https:\u002F\u002Fgithub.com\u002Fhuggingface\u002Ftransformers) library, you can easily use any of these models for training, evaluation, or inference.\n\nModels prefixed with a checkmark (✅) have been thoroughly tested and validated by the Oumi community, with ready-to-use recipes available in the [configs\u002Frecipes](configs\u002Frecipes) directory.\n\n\u003Cdetails>\n\u003Csummary>📋 Click to see more supported models\u003C\u002Fsummary>\n\n#### Instruct Models\n\n| Model | Size | Paper | HF Hub  | License  | Open [^1] |\n|-------|------|-------|---------|----------|------|\n| ✅ SmolLM-Instruct | 135M\u002F360M\u002F1.7B | [Blog](https:\u002F\u002Fhuggingface.co\u002Fblog\u002Fsmollm) | [Hub](https:\u002F\u002Fhuggingface.co\u002FHuggingFaceTB\u002FSmolLM-135M-Instruct) | Apache 2.0 | ✅ |\n| ✅ DeepSeek R1 Family | 1.5B\u002F8B\u002F32B\u002F70B\u002F671B | [Blog](https:\u002F\u002Fapi-docs.deepseek.com\u002Fnews\u002Fnews250120) | [Hub](https:\u002F\u002Fhuggingface.co\u002Fdeepseek-ai\u002FDeepSeek-R1) | MIT | ❌ |\n| ✅ Llama 3.1 Instruct | 8B\u002F70B\u002F405B | [Paper](https:\u002F\u002Farxiv.org\u002Fabs\u002F2407.21783) | [Hub](https:\u002F\u002Fhuggingface.co\u002Fmeta-llama\u002FLlama-3.1-70b-instruct) | [License](https:\u002F\u002Fllama.meta.com\u002Fllama3\u002Flicense\u002F) | ❌  |\n| ✅ Llama 3.2 Instruct | 1B\u002F3B | [Paper](https:\u002F\u002Farxiv.org\u002Fabs\u002F2407.21783) | [Hub](https:\u002F\u002Fhuggingface.co\u002Fmeta-llama\u002FLlama-3.2-3b-instruct) | [License](https:\u002F\u002Fllama.meta.com\u002Fllama3\u002Flicense\u002F) | ❌  |\n| ✅ Llama 3.3 Instruct | 70B | [Paper](https:\u002F\u002Farxiv.org\u002Fabs\u002F2407.21783) | [Hub](https:\u002F\u002Fhuggingface.co\u002Fmeta-llama\u002FLlama-3.3-70b-instruct) | [License](https:\u002F\u002Fllama.meta.com\u002Fllama3\u002Flicense\u002F) | ❌  |\n| ✅ Phi-3.5-Instruct | 4B\u002F14B | [Paper](https:\u002F\u002Farxiv.org\u002Fabs\u002F2404.14219) | [Hub](https:\u002F\u002Fhuggingface.co\u002Fmicrosoft\u002FPhi-3.5-mini-instruct) | [License](https:\u002F\u002Fhuggingface.co\u002Fmicrosoft\u002FPhi-3-mini-4k-instruct\u002Fblob\u002Fmain\u002FLICENSE) | ❌  |\n| ✅ Qwen3 | 0.6B-32B | [Paper](https:\u002F\u002Farxiv.org\u002Fabs\u002F2505.09388) | [Hub](https:\u002F\u002Fhuggingface.co\u002FQwen\u002FQwen3-32B) | [License](https:\u002F\u002Fgithub.com\u002FQwenLM\u002FQwen\u002Fblob\u002Fmain\u002FLICENSE) | ❌  |\n| Qwen2.5-Instruct | 0.5B-70B | [Paper](https:\u002F\u002Farxiv.org\u002Fabs\u002F2309.16609) | [Hub](https:\u002F\u002Fhuggingface.co\u002FQwen\u002FQwen2.5-7B-Instruct) | [License](https:\u002F\u002Fgithub.com\u002FQwenLM\u002FQwen\u002Fblob\u002Fmain\u002FLICENSE) | ❌  |\n| OLMo 2 Instruct | 7B | [Paper](https:\u002F\u002Farxiv.org\u002Fabs\u002F2402.00838) | [Hub](https:\u002F\u002Fhuggingface.co\u002Fallenai\u002FOLMo-2-1124-7B) | Apache 2.0 | ✅ |\n| ✅ OLMo 3 Instruct | 7B\u002F32B | [Paper](https:\u002F\u002Farxiv.org\u002Fabs\u002F2402.00838) | [Hub](https:\u002F\u002Fhuggingface.co\u002Fallenai\u002FOLMo-3-7B-Instruct) | Apache 2.0 | ✅ |\n| MPT-Instruct | 7B | [Blog](https:\u002F\u002Fwww.mosaicml.com\u002Fblog\u002Fmpt-7b) | [Hub](https:\u002F\u002Fhuggingface.co\u002Fmosaicml\u002Fmpt-7b-instruct) | Apache 2.0 | ✅ |\n| Command R | 35B\u002F104B | [Blog](https:\u002F\u002Fcohere.com\u002Fblog\u002Fcommand-r7b) | [Hub](https:\u002F\u002Fhuggingface.co\u002FCohereForAI\u002Fc4ai-command-r-plus) | [License](https:\u002F\u002Fcohere.com\u002Fc4ai-cc-by-nc-license) | ❌ |\n| Granite-3.1-Instruct | 2B\u002F8B | [Paper](https:\u002F\u002Fgithub.com\u002Fibm-granite\u002Fgranite-3.0-language-models\u002Fblob\u002Fmain\u002Fpaper.pdf) | [Hub](https:\u002F\u002Fhuggingface.co\u002Fibm-granite\u002Fgranite-3.1-8b-instruct) | Apache 2.0 | ❌ |\n| Gemma 2 Instruct | 2B\u002F9B | [Blog](https:\u002F\u002Fai.google.dev\u002Fgemma) | [Hub](https:\u002F\u002Fhuggingface.co\u002Fgoogle\u002Fgemma-2-2b-it) | [License](https:\u002F\u002Fai.google.dev\u002Fgemma\u002Fterms) | ❌ |\n| ✅ Gemma 3 Instruct | 4B\u002F12B\u002F27B | [Blog](https:\u002F\u002Fai.google.dev\u002Fgemma) | [Hub](https:\u002F\u002Fhuggingface.co\u002Fgoogle\u002Fgemma-3-27b-it) | [License](https:\u002F\u002Fai.google.dev\u002Fgemma\u002Fterms) | ❌ |\n| DBRX-Instruct | 130B MoE | [Blog](https:\u002F\u002Fwww.databricks.com\u002Fblog\u002Fintroducing-dbrx-new-state-art-open-llm) | [Hub](https:\u002F\u002Fhuggingface.co\u002Fdatabricks\u002Fdbrx-instruct) | Apache 2.0 | ❌ |\n| Falcon-Instruct | 7B\u002F40B | [Paper](https:\u002F\u002Farxiv.org\u002Fabs\u002F2306.01116) | [Hub](https:\u002F\u002Fhuggingface.co\u002Ftiiuae\u002Ffalcon-7b-instruct) | Apache 2.0 | ❌  |\n| ✅ Llama 4 Scout Instruct | 17B (Activated) 109B (Total) | [Paper](https:\u002F\u002Farxiv.org\u002Fabs\u002F2407.21783) | [Hub](https:\u002F\u002Fhuggingface.co\u002Fmeta-llama\u002FLlama-4-Scout-17B-16E-Instruct) | [License](https:\u002F\u002Fllama.meta.com\u002Fllama4\u002Flicense\u002F) | ❌  |\n| ✅ Llama 4 Maverick Instruct | 17B (Activated) 400B (Total) | [Paper](https:\u002F\u002Farxiv.org\u002Fabs\u002F2407.21783) | [Hub](https:\u002F\u002Fhuggingface.co\u002Fmeta-llama\u002FLlama-4-Maverick-17B-128E-Instruct) | [License](https:\u002F\u002Fllama.meta.com\u002Fllama4\u002Flicense\u002F) | ❌  |\n\n#### Vision-Language Models\n\n| Model | Size | Paper | HF Hub | License | Open |\n|-------|------|-------|---------|----------|------|\n| ✅ Llama 3.2 Vision | 11B | [Paper](https:\u002F\u002Farxiv.org\u002Fabs\u002F2407.21783) | [Hub](https:\u002F\u002Fhuggingface.co\u002Fmeta-llama\u002FLlama-3.2-11b-vision) | [License](https:\u002F\u002Fllama.meta.com\u002Fllama3\u002Flicense\u002F) | ❌  |\n| ✅ LLaVA-1.5 | 7B | [Paper](https:\u002F\u002Farxiv.org\u002Fabs\u002F2310.03744) | [Hub](https:\u002F\u002Fhuggingface.co\u002Fllava-hf\u002Fllava-1.5-7b-hf) | [License](https:\u002F\u002Fai.meta.com\u002Fllama\u002Flicense) | ❌ |\n| ✅ Phi-3 Vision | 4.2B | [Paper](https:\u002F\u002Farxiv.org\u002Fabs\u002F2404.14219) | [Hub](https:\u002F\u002Fhuggingface.co\u002Fmicrosoft\u002FPhi-3-vision-128k-instruct) | [License](https:\u002F\u002Fhuggingface.co\u002Fmicrosoft\u002FPhi-3-mini-4k-instruct\u002Fblob\u002Fmain\u002FLICENSE) | ❌ |\n| ✅ BLIP-2 | 3.6B | [Paper](https:\u002F\u002Farxiv.org\u002Fabs\u002F2301.12597) | [Hub](https:\u002F\u002Fhuggingface.co\u002FSalesforce\u002Fblip2-opt-2.7b) | MIT | ❌ |\n| ✅ Qwen2-VL | 2B | [Blog](https:\u002F\u002Fqwenlm.github.io\u002Fblog\u002Fqwen2-vl\u002F) | [Hub](https:\u002F\u002Fhuggingface.co\u002FQwen\u002FQwen2-VL-2B-Instruct) | [License](https:\u002F\u002Fgithub.com\u002FQwenLM\u002FQwen\u002Fblob\u002Fmain\u002FLICENSE) | ❌  |\n| ✅ Qwen3-VL | 2B\u002F4B\u002F8B | [Blog](https:\u002F\u002Fqwenlm.github.io\u002Fblog\u002Fqwen3-vl\u002F) | [Hub](https:\u002F\u002Fhuggingface.co\u002FQwen\u002FQwen3-VL-8B-Instruct) | [License](https:\u002F\u002Fgithub.com\u002FQwenLM\u002FQwen\u002Fblob\u002Fmain\u002FLICENSE) | ❌  |\n| ✅ SmolVLM-Instruct | 2B | [Blog](https:\u002F\u002Fhuggingface.co\u002Fblog\u002Fsmolvlm) | [Hub](https:\u002F\u002Fhuggingface.co\u002FHuggingFaceTB\u002FSmolVLM-Instruct) | Apache 2.0 | ✅  |\n\n#### Base Models\n\n| Model | Size | Paper | HF Hub | License | Open |\n|-------|------|-------|---------|----------|------|\n| ✅ SmolLM2 | 135M\u002F360M\u002F1.7B | [Blog](https:\u002F\u002Fhuggingface.co\u002Fblog\u002Fsmollm) | [Hub](https:\u002F\u002Fhuggingface.co\u002FHuggingFaceTB\u002FSmolLM2-135M) | Apache 2.0 | ✅ |\n| ✅ Llama 3.2 | 1B\u002F3B | [Paper](https:\u002F\u002Farxiv.org\u002Fabs\u002F2407.21783) | [Hub](https:\u002F\u002Fhuggingface.co\u002Fmeta-llama\u002FLlama-3.2-3b) | [License](https:\u002F\u002Fllama.meta.com\u002Fllama3\u002Flicense\u002F) | ❌  |\n| ✅ Llama 3.1 | 8B\u002F70B\u002F405B | [Paper](https:\u002F\u002Farxiv.org\u002Fabs\u002F2407.21783) | [Hub](https:\u002F\u002Fhuggingface.co\u002Fmeta-llama\u002FLlama-3.1-70b) | [License](https:\u002F\u002Fllama.meta.com\u002Fllama3\u002Flicense\u002F) | ❌  |\n| ✅ GPT-2 | 124M-1.5B | [Paper](https:\u002F\u002Farxiv.org\u002Fabs\u002F2005.14165) | [Hub](https:\u002F\u002Fhuggingface.co\u002Fgpt2) | MIT | ✅ |\n| DeepSeek V2 | 7B\u002F13B | [Blog](https:\u002F\u002Fwww.deepseek.com\u002Fblogs\u002Fdeepseek-v2) | [Hub](https:\u002F\u002Fhuggingface.co\u002Fdeepseek-ai\u002Fdeepseek-llm-7b-v2) | [License](https:\u002F\u002Fgithub.com\u002Fdeepseek-ai\u002FDeepSeek-LLM\u002Fblob\u002Fmain\u002FLICENSE-MODEL) | ❌ |\n| Gemma2 | 2B\u002F9B | [Blog](https:\u002F\u002Fai.google.dev\u002Fgemma) | [Hub](https:\u002F\u002Fhuggingface.co\u002Fgoogle\u002Fgemma2-7b) | [License](https:\u002F\u002Fai.google.dev\u002Fgemma\u002Fterms) | ❌ |\n| GPT-J | 6B | [Blog](https:\u002F\u002Fwww.eleuther.ai\u002Fartifacts\u002Fgpt-j) | [Hub](https:\u002F\u002Fhuggingface.co\u002FEleutherAI\u002Fgpt-j-6b) | Apache 2.0 | ✅ |\n| GPT-NeoX | 20B | [Paper](https:\u002F\u002Farxiv.org\u002Fabs\u002F2204.06745) | [Hub](https:\u002F\u002Fhuggingface.co\u002FEleutherAI\u002Fgpt-neox-20b) | Apache 2.0 | ✅ |\n| Mistral | 7B | [Paper](https:\u002F\u002Farxiv.org\u002Fabs\u002F2310.06825) | [Hub](https:\u002F\u002Fhuggingface.co\u002Fmistralai\u002FMistral-7B-v0.1) | Apache 2.0 | ❌  |\n| Mixtral | 8x7B\u002F8x22B | [Blog](https:\u002F\u002Fmistral.ai\u002Fnews\u002Fmixtral-of-experts\u002F) | [Hub](https:\u002F\u002Fhuggingface.co\u002Fmistralai\u002FMixtral-8x7B-v0.1) | Apache 2.0 | ❌  |\n| MPT | 7B | [Blog](https:\u002F\u002Fwww.mosaicml.com\u002Fblog\u002Fmpt-7b) | [Hub](https:\u002F\u002Fhuggingface.co\u002Fmosaicml\u002Fmpt-7b) | Apache 2.0 | ✅ |\n| OLMo | 1B\u002F7B | [Paper](https:\u002F\u002Farxiv.org\u002Fabs\u002F2402.00838) | [Hub](https:\u002F\u002Fhuggingface.co\u002Fallenai\u002FOLMo-7B-hf) | Apache 2.0 | ✅ |\n| ✅ Llama 4 Scout | 17B (Activated) 109B (Total) | [Paper](https:\u002F\u002Farxiv.org\u002Fabs\u002F2407.21783) | [Hub](https:\u002F\u002Fhuggingface.co\u002Fmeta-llama\u002FLlama-4-Scout-17B-16E) | [License](https:\u002F\u002Fllama.meta.com\u002Fllama4\u002Flicense\u002F) | ❌  |\n\n#### Reasoning Models\n\n| Model | Size | Paper | HF Hub | License | Open |\n|-------|------|-------|---------|----------|------|\n| ✅ gpt-oss | 20B\u002F120B | [Paper](https:\u002F\u002Farxiv.org\u002Fabs\u002F2508.10925) | [Hub](https:\u002F\u002Fhuggingface.co\u002Fopenai\u002Fgpt-oss-120b) | Apache 2.0 | ❌  |\n| ✅ Qwen3 | 0.6B-32B | [Paper](https:\u002F\u002Farxiv.org\u002Fabs\u002F2505.09388) | [Hub](https:\u002F\u002Fhuggingface.co\u002FQwen\u002FQwen3-32B) | [License](https:\u002F\u002Fgithub.com\u002FQwenLM\u002FQwen\u002Fblob\u002Fmain\u002FLICENSE) | ❌  |\n| ✅ Qwen3-Next | 80B-A3B | [Blog](https:\u002F\u002Fqwenlm.github.io\u002Fblog\u002Fqwen3\u002F) | [Hub](https:\u002F\u002Fhuggingface.co\u002FQwen\u002FQwen3-Next-80B-A3B) | [License](https:\u002F\u002Fgithub.com\u002FQwenLM\u002FQwen\u002Fblob\u002Fmain\u002FLICENSE) | ❌  |\n| Qwen QwQ | 32B | [Blog](https:\u002F\u002Fqwenlm.github.io\u002Fblog\u002Fqwq-32b-preview\u002F) | [Hub](https:\u002F\u002Fhuggingface.co\u002FQwen\u002FQwQ-32B-Preview) | [License](https:\u002F\u002Fgithub.com\u002FQwenLM\u002FQwen\u002Fblob\u002Fmain\u002FLICENSE) | ❌ |\n\n#### Code Models\n\n| Model | Size | Paper | HF Hub | License | Open |\n|-------|------|-------|---------|----------|------|\n| ✅ Qwen2.5 Coder | 0.5B-32B | [Blog](https:\u002F\u002Fqwenlm.github.io\u002Fblog\u002Fqwen2.5\u002F) | [Hub](https:\u002F\u002Fhuggingface.co\u002FQwen\u002FQwen2.5-Coder-32B-Instruct) | [License](https:\u002F\u002Fgithub.com\u002FQwenLM\u002FQwen\u002Fblob\u002Fmain\u002FLICENSE) | ❌  |\n| DeepSeek Coder | 1.3B-33B | [Paper](https:\u002F\u002Farxiv.org\u002Fabs\u002F2401.02954) | [Hub](https:\u002F\u002Fhuggingface.co\u002Fdeepseek-ai\u002Fdeepseek-coder-7b-instruct) | [License](https:\u002F\u002Fgithub.com\u002Fdeepseek-ai\u002FDeepSeek-LLM\u002Fblob\u002Fmain\u002FLICENSE-MODEL) | ❌  |\n| StarCoder 2 | 3B\u002F7B\u002F15B | [Paper](https:\u002F\u002Farxiv.org\u002Fabs\u002F2402.19173) | [Hub](https:\u002F\u002Fhuggingface.co\u002Fbigcode\u002Fstarcoder2-15b) | [License](https:\u002F\u002Fhuggingface.co\u002Fspaces\u002Fbigcode\u002Fbigcode-model-license-agreement) | ✅ |\n\n#### Math Models\n\n| Model | Size | Paper | HF Hub | License | Open |\n|-------|------|-------|---------|----------|------|\n| DeepSeek Math | 7B | [Paper](https:\u002F\u002Farxiv.org\u002Fabs\u002F2401.02954) | [Hub](https:\u002F\u002Fhuggingface.co\u002Fdeepseek-ai\u002Fdeepseek-math-7b-instruct) | [License](https:\u002F\u002Fgithub.com\u002Fdeepseek-ai\u002FDeepSeek-LLM\u002Fblob\u002Fmain\u002FLICENSE-MODEL) | ❌  | |\n\n\u003C\u002Fdetails>\n\n## 📖 Documentation\n\nTo learn more about all the platform's capabilities, see the [Oumi documentation](https:\u002F\u002Foumi.ai\u002Fdocs).\n\n## 🤝 Join the Community\n\nOumi is a community-first effort. Whether you are a developer, a researcher, or a non-technical user, all contributions are very welcome!\n\n- To contribute to the `oumi` repository, please check the [`CONTRIBUTING.md`](https:\u002F\u002Fgithub.com\u002Foumi-ai\u002Foumi\u002Fblob\u002Fmain\u002FCONTRIBUTING.md) for guidance on how to contribute to send your first Pull Request.\n- Make sure to join our [Discord community](https:\u002F\u002Fdiscord.gg\u002Foumi) to get help, share your experiences, and contribute to the project!\n- If you are interested in joining one of the community's open-science efforts, check out our [open collaboration](https:\u002F\u002Foumi.ai\u002Fcommunity) page.\n\n## 🙏 Acknowledgements\n\nOumi makes use of [several libraries](https:\u002F\u002Foumi.ai\u002Fdocs\u002Fen\u002Flatest\u002Fabout\u002Facknowledgements.html) and tools from the open-source community. We would like to acknowledge and deeply thank the contributors of these projects! ✨ 🌟 💫\n\n## 📝 Citation\n\nIf you find Oumi useful in your research, please consider citing it:\n\n```bibtex\n@software{oumi2025,\n  author = {Oumi Community},\n  title = {Oumi: an Open, End-to-end Platform for Building Large Foundation Models},\n  month = {January},\n  year = {2025},\n  url = {https:\u002F\u002Fgithub.com\u002Foumi-ai\u002Foumi}\n}\n```\n\n## 📜 License\n\nThis project is licensed under the Apache License 2.0. See the [LICENSE](LICENSE) file for details.\n\n[^1]: Open models are defined as models with fully open weights, training code, and data, and a permissive license. See [Open Source Definitions](https:\u002F\u002Fopensource.org\u002Fai) for more information.\n","oumi-ai\u002Foumi 是一个用于轻松微调、评估和部署多种开源大语言模型（如Gemma 4, Qwen3.5, Qwen3.6, gpt-oss, DeepSeek-R1等）的工具。该项目支持DPO、SFT等微调方法，并提供了强大的评估与推理功能，能够帮助用户快速搭建从训练到部署的一站式解决方案。它特别适用于需要定制化调整现有预训练模型以满足特定需求的应用场景，例如构建企业级聊天机器人或专业领域的知识问答系统。此外，oumi还集成了对最新版本Transformers库的支持，确保了与最前沿技术的兼容性。",2,"2026-06-11 03:39:44","high_star"]