经 AI Skill Hub 精选评估,Oumi模型微调部署工具 获评「强烈推荐」。已获得 9.2k 颗 GitHub Star,这款AI工具在功能完整性、社区活跃度和易用性方面表现出色,AI 评分 8.2 分,适合有一定技术背景的用户使用。
Oumi模型微调部署工具 是一款基于 Python 开发的开源工具,专注于 模型微调、DPO训练、大模型评估 等核心功能。作为 GitHub 开源项目,它拥有活跃的社区支持和持续的版本迭代,代码完全透明可审计,支持本地部署以保护数据隐私。无论是个人使用还是集成到企业工作流,都能提供稳定可靠的解决方案。
Oumi模型微调部署工具 是一款基于 Python 开发的开源工具,专注于 模型微调、DPO训练、大模型评估 等核心功能。作为 GitHub 开源项目,它拥有活跃的社区支持和持续的版本迭代,代码完全透明可审计,支持本地部署以保护数据隐私。无论是个人使用还是集成到企业工作流,都能提供稳定可靠的解决方案。
# 方式一:pip 安装(推荐)
pip install oumi
# 方式二:虚拟环境安装(推荐生产环境)
python -m venv .venv
source .venv/bin/activate # Windows: .venv\Scripts\activate
pip install oumi
# 方式三:从源码安装(获取最新功能)
git clone https://github.com/oumi-ai/oumi
cd oumi
pip install -e .
# 验证安装
python -c "import oumi; print('安装成功')"
# 命令行使用
oumi --help
# 基本用法
oumi input_file -o output_file
# Python 代码中调用
import oumi
# 示例
result = oumi.process("input")
print(result)
# oumi 配置文件示例(config.yml) app: name: "oumi" debug: false log_level: "INFO" # 运行时指定配置文件 oumi --config config.yml # 或通过环境变量配置 export OUMI_API_KEY="your-key" export OUMI_OUTPUT_DIR="./output"
Oumi 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.
With Oumi, you can:
All with one consistent API, production-grade reliability, and all the flexibility you need for research.
Learn more at oumi.ai, or jump right in with the quickstart guide.
<p align="center"> <a href="https://trendshift.io/repositories/12865"> <img alt="GitHub trending" src="https://trendshift.io/api/badge/repositories/12865" /> </a> </p>
| **Notebook** | **Try in Colab** | **Goal** |
|---|---|---|
| **🎯 Getting Started: A Tour** | <a target="_blank" href="https://colab.research.google.com/github/oumi-ai/oumi/blob/main/notebooks/Oumi - A Tour.ipynb"><img src="https://colab.research.google.com/assets/colab-badge.svg" alt="Open In Colab"/></a> | Quick tour of core features: training, evaluation, inference, and job management |
| **🔧 Model Finetuning Guide** | <a target="_blank" href="https://colab.research.google.com/github/oumi-ai/oumi/blob/main/notebooks/Oumi - Finetuning Tutorial.ipynb"><img src="https://colab.research.google.com/assets/colab-badge.svg" alt="Open In Colab"/></a> | End-to-end guide to LoRA tuning with data prep, training, and evaluation |
| **📚 Model Distillation** | <a target="_blank" href="https://colab.research.google.com/github/oumi-ai/oumi/blob/main/notebooks/Oumi - Distill a Large Model.ipynb"><img src="https://colab.research.google.com/assets/colab-badge.svg" alt="Open In Colab"/></a> | Guide to distilling large models into smaller, efficient ones |
| **📋 Model Evaluation** | <a target="_blank" href="https://colab.research.google.com/github/oumi-ai/oumi/blob/main/notebooks/Oumi - Evaluation with Oumi.ipynb"><img src="https://colab.research.google.com/assets/colab-badge.svg" alt="Open In Colab"/></a> | Comprehensive model evaluation using Oumi's evaluation framework |
| **☁️ Remote Training** | <a target="_blank" href="https://colab.research.google.com/github/oumi-ai/oumi/blob/main/notebooks/Oumi - Running Jobs Remotely.ipynb"><img src="https://colab.research.google.com/assets/colab-badge.svg" alt="Open In Colab"/></a> | Launch and monitor training jobs on cloud (AWS, Azure, GCP, Lambda, etc.) platforms |
| **📈 LLM-as-a-Judge** | <a target="_blank" href="https://colab.research.google.com/github/oumi-ai/oumi/blob/main/notebooks/Oumi - Simple Judge.ipynb"><img src="https://colab.research.google.com/assets/colab-badge.svg" alt="Open In Colab"/></a> | Filter and curate training data with built-in judges |
Choose the installation method that works best for you:
<details open> <summary><b>Using pip (Recommended)</b></summary>
```bash
uv pip install oumi
Explore the growing collection of ready-to-use configurations for state-of-the-art models and training workflows:
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, and datasets (supervised fine-tuning, pre-training, preference tuning, and vision-language finetuning) in the oumi documentation.
docker run --gpus all -v $(pwd):/workspace -it ghcr.io/oumi-ai/oumi:latest \ oumi train --config /workspace/my_config.yaml
</details>
<details>
<summary><b>Quick Install Script (Experimental)</b></summary>
Try Oumi without setting up a Python environment. This installs Oumi in an isolated environment:
bash curl -LsSf https://oumi.ai/install.sh | bash ```
</details>
For more advanced installation options, see the installation guide.
This section lists all the language models that can be used with Oumi. Thanks to the integration with the 🤗 Transformers library, you can easily use any of these models for training, evaluation, or inference.
Models prefixed with a checkmark (✅) have been thoroughly tested and validated by the Oumi community, with ready-to-use recipes available in the configs/recipes directory.
<details> <summary>📋 Click to see more supported models</summary>
| Model | Size | Paper | HF Hub | License | Open [^1] |
|---|---|---|---|---|---|
| ✅ SmolLM-Instruct | 135M/360M/1.7B | [Blog](https://huggingface.co/blog/smollm) | [Hub](https://huggingface.co/HuggingFaceTB/SmolLM-135M-Instruct) | Apache 2.0 | ✅ |
| ✅ DeepSeek R1 Family | 1.5B/8B/32B/70B/671B | [Blog](https://api-docs.deepseek.com/news/news250120) | [Hub](https://huggingface.co/deepseek-ai/DeepSeek-R1) | MIT | ❌ |
| ✅ Llama 3.1 Instruct | 8B/70B/405B | [Paper](https://arxiv.org/abs/2407.21783) | [Hub](https://huggingface.co/meta-llama/Llama-3.1-70b-instruct) | [License](https://llama.meta.com/llama3/license/) | ❌ |
| ✅ Llama 3.2 Instruct | 1B/3B | [Paper](https://arxiv.org/abs/2407.21783) | [Hub](https://huggingface.co/meta-llama/Llama-3.2-3b-instruct) | [License](https://llama.meta.com/llama3/license/) | ❌ |
| ✅ Llama 3.3 Instruct | 70B | [Paper](https://arxiv.org/abs/2407.21783) | [Hub](https://huggingface.co/meta-llama/Llama-3.3-70b-instruct) | [License](https://llama.meta.com/llama3/license/) | ❌ |
| ✅ Phi-3.5-Instruct | 4B/14B | [Paper](https://arxiv.org/abs/2404.14219) | [Hub](https://huggingface.co/microsoft/Phi-3.5-mini-instruct) | [License](https://huggingface.co/microsoft/Phi-3-mini-4k-instruct/blob/main/LICENSE) | ❌ |
| ✅ Qwen3 | 0.6B-32B | [Paper](https://arxiv.org/abs/2505.09388) | [Hub](https://huggingface.co/Qwen/Qwen3-32B) | [License](https://github.com/QwenLM/Qwen/blob/main/LICENSE) | ❌ |
| Qwen2.5-Instruct | 0.5B-70B | [Paper](https://arxiv.org/abs/2309.16609) | [Hub](https://huggingface.co/Qwen/Qwen2.5-7B-Instruct) | [License](https://github.com/QwenLM/Qwen/blob/main/LICENSE) | ❌ |
| OLMo 2 Instruct | 7B | [Paper](https://arxiv.org/abs/2402.00838) | [Hub](https://huggingface.co/allenai/OLMo-2-1124-7B) | Apache 2.0 | ✅ |
| ✅ OLMo 3 Instruct | 7B/32B | [Paper](https://arxiv.org/abs/2402.00838) | [Hub](https://huggingface.co/allenai/OLMo-3-7B-Instruct) | Apache 2.0 | ✅ |
| MPT-Instruct | 7B | [Blog](https://www.mosaicml.com/blog/mpt-7b) | [Hub](https://huggingface.co/mosaicml/mpt-7b-instruct) | Apache 2.0 | ✅ |
| Command R | 35B/104B | [Blog](https://cohere.com/blog/command-r7b) | [Hub](https://huggingface.co/CohereForAI/c4ai-command-r-plus) | [License](https://cohere.com/c4ai-cc-by-nc-license) | ❌ |
| Granite-3.1-Instruct | 2B/8B | [Paper](https://github.com/ibm-granite/granite-3.0-language-models/blob/main/paper.pdf) | [Hub](https://huggingface.co/ibm-granite/granite-3.1-8b-instruct) | Apache 2.0 | ❌ |
| Gemma 2 Instruct | 2B/9B | [Blog](https://ai.google.dev/gemma) | [Hub](https://huggingface.co/google/gemma-2-2b-it) | [License](https://ai.google.dev/gemma/terms) | ❌ |
| ✅ Gemma 3 Instruct | 4B/12B/27B | [Blog](https://ai.google.dev/gemma) | [Hub](https://huggingface.co/google/gemma-3-27b-it) | [License](https://ai.google.dev/gemma/terms) | ❌ |
| DBRX-Instruct | 130B MoE | [Blog](https://www.databricks.com/blog/introducing-dbrx-new-state-art-open-llm) | [Hub](https://huggingface.co/databricks/dbrx-instruct) | Apache 2.0 | ❌ |
| Falcon-Instruct | 7B/40B | [Paper](https://arxiv.org/abs/2306.01116) | [Hub](https://huggingface.co/tiiuae/falcon-7b-instruct) | Apache 2.0 | ❌ |
| ✅ Llama 4 Scout Instruct | 17B (Activated) 109B (Total) | [Paper](https://arxiv.org/abs/2407.21783) | [Hub](https://huggingface.co/meta-llama/Llama-4-Scout-17B-16E-Instruct) | [License](https://llama.meta.com/llama4/license/) | ❌ |
| ✅ Llama 4 Maverick Instruct | 17B (Activated) 400B (Total) | [Paper](https://arxiv.org/abs/2407.21783) | [Hub](https://huggingface.co/meta-llama/Llama-4-Maverick-17B-128E-Instruct) | [License](https://llama.meta.com/llama4/license/) | ❌ |
| Model | Size | Paper | HF Hub | License | Open |
|---|---|---|---|---|---|
| ✅ Llama 3.2 Vision | 11B | [Paper](https://arxiv.org/abs/2407.21783) | [Hub](https://huggingface.co/meta-llama/Llama-3.2-11b-vision) | [License](https://llama.meta.com/llama3/license/) | ❌ |
| ✅ LLaVA-1.5 | 7B | [Paper](https://arxiv.org/abs/2310.03744) | [Hub](https://huggingface.co/llava-hf/llava-1.5-7b-hf) | [License](https://ai.meta.com/llama/license) | ❌ |
| ✅ Phi-3 Vision | 4.2B | [Paper](https://arxiv.org/abs/2404.14219) | [Hub](https://huggingface.co/microsoft/Phi-3-vision-128k-instruct) | [License](https://huggingface.co/microsoft/Phi-3-mini-4k-instruct/blob/main/LICENSE) | ❌ |
| ✅ BLIP-2 | 3.6B | [Paper](https://arxiv.org/abs/2301.12597) | [Hub](https://huggingface.co/Salesforce/blip2-opt-2.7b) | MIT | ❌ |
| ✅ Qwen2-VL | 2B | [Blog](https://qwenlm.github.io/blog/qwen2-vl/) | [Hub](https://huggingface.co/Qwen/Qwen2-VL-2B-Instruct) | [License](https://github.com/QwenLM/Qwen/blob/main/LICENSE) | ❌ |
| ✅ Qwen3-VL | 2B/4B/8B | [Blog](https://qwenlm.github.io/blog/qwen3-vl/) | [Hub](https://huggingface.co/Qwen/Qwen3-VL-8B-Instruct) | [License](https://github.com/QwenLM/Qwen/blob/main/LICENSE) | ❌ |
| ✅ SmolVLM-Instruct | 2B | [Blog](https://huggingface.co/blog/smolvlm) | [Hub](https://huggingface.co/HuggingFaceTB/SmolVLM-Instruct) | Apache 2.0 | ✅ |
| Model | Size | Paper | HF Hub | License | Open |
|---|---|---|---|---|---|
| ✅ SmolLM2 | 135M/360M/1.7B | [Blog](https://huggingface.co/blog/smollm) | [Hub](https://huggingface.co/HuggingFaceTB/SmolLM2-135M) | Apache 2.0 | ✅ |
| ✅ Llama 3.2 | 1B/3B | [Paper](https://arxiv.org/abs/2407.21783) | [Hub](https://huggingface.co/meta-llama/Llama-3.2-3b) | [License](https://llama.meta.com/llama3/license/) | ❌ |
| ✅ Llama 3.1 | 8B/70B/405B | [Paper](https://arxiv.org/abs/2407.21783) | [Hub](https://huggingface.co/meta-llama/Llama-3.1-70b) | [License](https://llama.meta.com/llama3/license/) | ❌ |
| ✅ GPT-2 | 124M-1.5B | [Paper](https://arxiv.org/abs/2005.14165) | [Hub](https://huggingface.co/gpt2) | MIT | ✅ |
| DeepSeek V2 | 7B/13B | [Blog](https://www.deepseek.com/blogs/deepseek-v2) | [Hub](https://huggingface.co/deepseek-ai/deepseek-llm-7b-v2) | [License](https://github.com/deepseek-ai/DeepSeek-LLM/blob/main/LICENSE-MODEL) | ❌ |
| Gemma2 | 2B/9B | [Blog](https://ai.google.dev/gemma) | [Hub](https://huggingface.co/google/gemma2-7b) | [License](https://ai.google.dev/gemma/terms) | ❌ |
| GPT-J | 6B | [Blog](https://www.eleuther.ai/artifacts/gpt-j) | [Hub](https://huggingface.co/EleutherAI/gpt-j-6b) | Apache 2.0 | ✅ |
| GPT-NeoX | 20B | [Paper](https://arxiv.org/abs/2204.06745) | [Hub](https://huggingface.co/EleutherAI/gpt-neox-20b) | Apache 2.0 | ✅ |
| Mistral | 7B | [Paper](https://arxiv.org/abs/2310.06825) | [Hub](https://huggingface.co/mistralai/Mistral-7B-v0.1) | Apache 2.0 | ❌ |
| Mixtral | 8x7B/8x22B | [Blog](https://mistral.ai/news/mixtral-of-experts/) | [Hub](https://huggingface.co/mistralai/Mixtral-8x7B-v0.1) | Apache 2.0 | ❌ |
| MPT | 7B | [Blog](https://www.mosaicml.com/blog/mpt-7b) | [Hub](https://huggingface.co/mosaicml/mpt-7b) | Apache 2.0 | ✅ |
| OLMo | 1B/7B | [Paper](https://arxiv.org/abs/2402.00838) | [Hub](https://huggingface.co/allenai/OLMo-7B-hf) | Apache 2.0 | ✅ |
| ✅ Llama 4 Scout | 17B (Activated) 109B (Total) | [Paper](https://arxiv.org/abs/2407.21783) | [Hub](https://huggingface.co/meta-llama/Llama-4-Scout-17B-16E) | [License](https://llama.meta.com/llama4/license/) | ❌ |
| Model | Size | Paper | HF Hub | License | Open |
|---|---|---|---|---|---|
| ✅ gpt-oss | 20B/120B | [Paper](https://arxiv.org/abs/2508.10925) | [Hub](https://huggingface.co/openai/gpt-oss-120b) | Apache 2.0 | ❌ |
| ✅ Qwen3 | 0.6B-32B | [Paper](https://arxiv.org/abs/2505.09388) | [Hub](https://huggingface.co/Qwen/Qwen3-32B) | [License](https://github.com/QwenLM/Qwen/blob/main/LICENSE) | ❌ |
| ✅ Qwen3-Next | 80B-A3B | [Blog](https://qwenlm.github.io/blog/qwen3/) | [Hub](https://huggingface.co/Qwen/Qwen3-Next-80B-A3B) | [License](https://github.com/QwenLM/Qwen/blob/main/LICENSE) | ❌ |
| Qwen QwQ | 32B | [Blog](https://qwenlm.github.io/blog/qwq-32b-preview/) | [Hub](https://huggingface.co/Qwen/QwQ-32B-Preview) | [License](https://github.com/QwenLM/Qwen/blob/main/LICENSE) | ❌ |
| Model | Size | Paper | HF Hub | License | Open |
|---|---|---|---|---|---|
| ✅ Qwen2.5 Coder | 0.5B-32B | [Blog](https://qwenlm.github.io/blog/qwen2.5/) | [Hub](https://huggingface.co/Qwen/Qwen2.5-Coder-32B-Instruct) | [License](https://github.com/QwenLM/Qwen/blob/main/LICENSE) | ❌ |
| DeepSeek Coder | 1.3B-33B | [Paper](https://arxiv.org/abs/2401.02954) | [Hub](https://huggingface.co/deepseek-ai/deepseek-coder-7b-instruct) | [License](https://github.com/deepseek-ai/DeepSeek-LLM/blob/main/LICENSE-MODEL) | ❌ |
| StarCoder 2 | 3B/7B/15B | [Paper](https://arxiv.org/abs/2402.19173) | [Hub](https://huggingface.co/bigcode/starcoder2-15b) | [License](https://huggingface.co/spaces/bigcode/bigcode-model-license-agreement) | ✅ |
| Model | Size | Paper | HF Hub | License | Open | |
|---|---|---|---|---|---|---|
| DeepSeek Math | 7B | [Paper](https://arxiv.org/abs/2401.02954) | [Hub](https://huggingface.co/deepseek-ai/deepseek-math-7b-instruct) | [License](https://github.com/deepseek-ai/DeepSeek-LLM/blob/main/LICENSE-MODEL) | ❌ |
</details>
高质量开源微调框架,集成DPO/评估/部署全流程,维护活跃,Star增长快,适合专业开发者和研究机构使用。
AI Skill Hub 为第三方内容聚合平台,本页面信息基于公开数据整理,不对工具功能和质量作任何法律背书。
建议在沙箱或测试环境中充分验证后,再部署至生产环境,并做好必要的安全评估。
✅ Apache 2.0 — 宽松开源协议,可商用,需保留版权声明和 NOTICE 文件,含专利授权条款。
AI Skill Hub 点评:Oumi模型微调部署工具 的核心功能完整,质量优秀。对于AI 技术爱好者来说,这是一个值得纳入个人工具库的选择。建议先在非生产环境试用,再逐步推广。
| 原始名称 | oumi |
| 原始描述 | 开源AI工具:Easily fine-tune, evaluate and deploy Gemma 4, Qwen3.5, Qwen3.6, gpt-oss, DeepSe。⭐9.2k · Python |
| Topics | 模型微调DPO训练大模型评估模型部署开源框架 |
| GitHub | https://github.com/oumi-ai/oumi |
| License | Apache-2.0 |
| 语言 | Python |
收录时间:2026-05-19 · 更新时间:2026-05-30 · License:Apache-2.0 · AI Skill Hub 不对第三方内容的准确性作法律背书。