orca2

Orca 2 由微软研究院开发，是 Meta Llama 2 模型的微调版本。该模型旨在特别擅长推理。

7b 13b

56.7K 拉取更新 11 个月前

33 个标签

11 个月前更新

11 个月前

f0dd7ec070fa · 2.9GB

{"num_ctx":4096,"stop":["\u003c|im_start","\u003c|im_end","|im_start","|im_end"]}

82B

模板

<|im_start|>system {{ .System }}<|im_end|> <|im_start|>user {{ .Prompt }}<|im_end|> <|im_start|>assi

106B

系统

你是 Orca，一个由微软创建的 AI 语言模型。你是一个谨慎的助手。你仔细

215B

许可证

微软研究许可条款如果你住在美国，请阅读“约束性仲裁”

11kB

自述文件

Orca 2 模型由微软研究院构建。它们使用合成数据集对 Meta 的 Llama 2 进行微调，该数据集旨在增强小型模型的推理能力。所有合成训练数据都使用微软 Azure 内容过滤器进行审核。

用例

Orca 2 是一个有用的助手，它在推理你给定的数据、阅读理解、数学问题解决和文本摘要等任务中提供答案。

该模型旨在特别擅长推理。

微软研究院希望这款模型能够鼓励对更小的语言模型的开发、评估和对齐进行进一步的研究。

CLI

70 亿参数模型

ollama run orca2

130 亿参数模型

ollama run orca2:13b

API

示例

curl -X POST https://127.0.0.1:11434/api/generate -d '{
  "model": "orca2",
  "prompt":"Why is the sky blue?"
 }'

参考

微软研究院公告博客
微软研究院发表的论文，Orca 2：教小型语言模型如何推理

Orca 2 models are built by Microsoft Research. They are fine-tuned on Meta's Llama 2 using a synthetic dataset that was created to enhance the small model's reasoning abilities. All synthetic training data was moderated using Microsoft Azure content filters.

### Use case

Orca 2 is a helpful assistant, and provides an answer in tasks such as reasoning over your given data, reading comprehension, math problem solving and text summarization.

The model is designed to excel particularly in reasoning.

Microsoft Research's intended purpose for this model is to encourage further research on the development, evaluation, and alignment of smaller language models.

### CLI

7 billion parameter model:

```
ollama run orca2
```

13 billion parameter model:

```
ollama run orca2:13b
```

### API

Example:

```bash
curl -X POST https://127.0.0.1:11434/api/generate -d '{
  "model": "orca2",
  "prompt":"Why is the sky blue?"
 }'
```

### References

- [Microsoft Research announcement blog](https://www.microsoft.com/en-us/research/blog/orca-2-teaching-small-language-models-how-to-reason/)

- Microsoft Research published paper, [Orca 2: Teaching Small Language Models How to Reason](https://arxiv.org/pdf/2311.11045.pdf)

粘贴、拖放或点击上传图像（.png、.jpeg、.jpg、.svg、.gif）