|
--- |
|
license: apache-2.0 |
|
datasets: |
|
- Azure99/blossom-chat-v1 |
|
- Azure99/blossom-math-v1 |
|
- ehartford/dolphin |
|
- WizardLM/WizardLM_evol_instruct_V2_196k |
|
language: |
|
- zh |
|
- en |
|
--- |
|
# **BLOSSOM-v2-3b** |
|
|
|
### 介绍 |
|
|
|
Blossom是一个对话式语言模型,基于Bloom-3b预训练模型,在Blossom、Wizard、Dolphin混合数据集上进行指令精调得来。 |
|
|
|
训练分为两阶段,第一阶段使用120K Wizard、180K Dolphin单轮指令数据集,训练1个epoch;第二阶段使用60K Blossom chat、2K Blossom math多轮对话数据集,训练3个epoch。 |
|
|
|
### 推理 |
|
|
|
推理采用对话续写的形式。 |
|
|
|
单轮对话 |
|
|
|
``` |
|
A chat between a human and an artificial intelligence bot. The bot gives helpful, detailed, and polite answers to the human's questions. |
|
|Human|: 你好 |
|
|Bot|: |
|
``` |
|
|
|
多轮对话 |
|
|
|
``` |
|
A chat between a human and an artificial intelligence bot. The bot gives helpful, detailed, and polite answers to the human's questions. |
|
|Human|: 你好 |
|
|Bot|: 你好,有什么我能帮助你的?</s> |
|
|Human|: 介绍下中国的首都吧 |
|
|Bot|: |
|
``` |
|
|
|
注意:在历史对话的Bot输出结尾,拼接一个</s> |
|
# [Open LLM Leaderboard Evaluation Results](https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard) |
|
Detailed results can be found [here](https://huggingface.co/datasets/open-llm-leaderboard/details_Azure99__blossom-v2-3b) |
|
|
|
| Metric | Value | |
|
|-----------------------|---------------------------| |
|
| Avg. | 32.43 | |
|
| ARC (25-shot) | 35.32 | |
|
| HellaSwag (10-shot) | 54.1 | |
|
| MMLU (5-shot) | 23.99 | |
|
| TruthfulQA (0-shot) | 43.11 | |
|
| Winogrande (5-shot) | 58.8 | |
|
| GSM8K (5-shot) | 0.53 | |
|
| DROP (3-shot) | 11.17 | |
|
|