tulu-v3.1-mix-preview-4096-OLMoE

# OLMoE SFT Mix

The SFT mix used is an expanded version of the [Tulu v2 SFT mix](https://huggingface.co/datasets/allenai/tulu-v2-sft-mixture-olmo-4096) with new additions for code, [CodeFeedback-Filtered-Instruction](https://huggingface.co/datasets/m-a-p/CodeFeedback-Filtered-Instruction), reasoning, [MetaMathQA](https://huggingface.co/datasets/meta-math/MetaMathQA), and instruction following, [No Robots](https://huggingface.co/datasets/HuggingFaceH4/no_robots) and a subset of [Daring Anteater](https://huggingface.co/datasets/nvidia/Daring-Anteater).

Please see the referenced datasets for the multiple licenses used in subsequent data.
We do not introduce any new data with this dataset.
Config for creation via [`open-instruct`](https://github.com/allenai/open-instruct/blob/main/open_instruct/mix_data.py):
```
dataset_mixer:
allenai/tulu-v2-sft-mixture-olmo-4096: 1.0
HuggingFaceH4/no_robots: 1.0
meta-math/MetaMathQA: 0.25
m-a-p/CodeFeedback-Filtered-Instruction: 1.0
ai2-adapt-dev/daring-anteater-specialized: 1.0
max_seq_length: 4096
```

Reanming code:
```
def rename_messages(example):
messages = example["messages"]
new_messages = []
for m in messages:
new_messages.append({"role": m["role"], "content":m["content"].replace("OLMo","OLMoE")})
example["messages"] = new_messages
return example
```

Related datasets (for updated list, see [collection](https://huggingface.co/collections/allenai/tulu-3-data-mixes-66a944d48990fafa62c2c18c))
| Version | Name | Summary | Max Length | Model Name |
|---------|------|---------|------------|------------|
| v1 | [allenai/tulu-v1-sft-mixture](https://huggingface.co/datasets/allenai/tulu-v1-sft-mixture) | | | |
| v2 | [allenai/tulu-v2-sft-mixture](https://huggingface.co/datasets/allenai/tulu-v2-sft-mixture) | | - | |
| v2 | [allenai/tulu-v2-sft-mixture-olmo-2048](https://huggingface.co/datasets/allenai/tulu-v2-sft-mixture-olmo-2048) | | 2048 | OLMo-2048 |
| v3.0 | [allenai/tulu-v3.0-mix-preview-4096-OLMo](https://huggingface.co/datasets/allenai/tulu-v3.0-mix-preview-4096-OLMo) | Tulu 2 + Math/Code + No Robots| 4096 | OLMo |
| v3.0 | [allenai/tulu-v3.0-mix-preview-4096-OLMoE](https://huggingface.co/datasets/allenai/tulu-v3.0-mix-preview-4096-OLMoE) | OLMoE Name| 4096 | OLMoE |
| v3.1 | [**allenai/tulu-v3.1-mix-preview-4096-OLMoE**](https://huggingface.co/datasets/allenai/tulu-v3.1-mix-preview-4096-OLMoE) | Add specialized Daring Anteater | 4096 | OLMoE |

# OLMoE 监督微调（Supervised Fine-Tuning，SFT）混合数据集

本次使用的SFT混合数据集是[Tulu v2 SFT混合数据集](https://huggingface.co/datasets/allenai/tulu-v2-sft-mixture-olmo-4096)的扩展版本，新增了代码相关数据集[CodeFeedback-Filtered-Instruction](https://huggingface.co/datasets/m-a-p/CodeFeedback-Filtered-Instruction)、推理相关数据集[MetaMathQA](https://huggingface.co/datasets/meta-math/MetaMathQA)、指令遵循数据集[No Robots](https://huggingface.co/datasets/HuggingFaceH4/no_robots)，以及[Daring Anteater](https://huggingface.co/datasets/nvidia/Daring-Anteater)的子集。

有关本数据集后续所使用的多种许可协议，请参阅所引用的各原始数据集。本数据集未引入任何全新的自有数据。

本数据集通过[`open-instruct`](https://github.com/allenai/open-instruct/blob/main/open_instruct/mix_data.py)工具生成，对应的配置参数如下：

dataset_mixer:
allenai/tulu-v2-sft-mixture-olmo-4096: 1.0
HuggingFaceH4/no_robots: 1.0
meta-math/MetaMathQA: 0.25
m-a-p/CodeFeedback-Filtered-Instruction: 1.0
ai2-adapt-dev/daring-anteater-specialized: 1.0
max_seq_length: 4096

以下为消息重命名处理代码：
python
def rename_messages(example):
messages = example["messages"]
new_messages = []
for m in messages:
new_messages.append({"role": m["role"], "content":m["content"].replace("OLMo","OLMoE")})
example["messages"] = new_messages
return example

相关数据集（完整列表请参阅[数据集集合](https://huggingface.co/collections/allenai/tulu-3-data-mixes-66a944d48990fafa62c2c18c)）：
| 版本号 | 名称 | 摘要 | 最大序列长度 | 模型名称 |
|---------|------|---------|------------|------------|
| v1 | [allenai/tulu-v1-sft-mixture](https://huggingface.co/datasets/allenai/tulu-v1-sft-mixture) | 无 | 无 | 无 |
| v2 | [allenai/tulu-v2-sft-mixture](https://huggingface.co/datasets/allenai/tulu-v2-sft-mixture) | 无 | - | 无 |
| v2 | [allenai/tulu-v2-sft-mixture-olmo-2048](https://huggingface.co/datasets/allenai/tulu-v2-sft-mixture-olmo-2048) | 无 | 2048 | OLMo-2048 |
| v3.0 | [allenai/tulu-v3.0-mix-preview-4096-OLMo](https://huggingface.co/datasets/allenai/tulu-v3.0-mix-preview-4096-OLMo) | Tulu 2 + 数学/代码 + No Robots | 4096 | OLMo |
| v3.0 | [allenai/tulu-v3.0-mix-preview-4096-OLMoE](https://huggingface.co/datasets/allenai/tulu-v3.0-mix-preview-4096-OLMoE) | 适配OLMoE命名 | 4096 | OLMoE |
| v3.1 | [**allenai/tulu-v3.1-mix-preview-4096-OLMoE**](https://huggingface.co/datasets/allenai/tulu-v3.1-mix-preview-4096-OLMoE) | 新增专用版Daring Anteater数据集 | 4096 | OLMoE |