| # internlm2.5_7b_distill | |
| ## 架构图 | |
| <div align="center"> | |
| <img src="distill.png" width="800"/> | |
| </div> | |
| ## 基座模型 | |
| https://huggingface.co/Slipstream-Max/Emollm-InternLM2.5-7B-chat-GGUF-fp16 | |
| ## 数据集 | |
| ### 数据集组成 | |
| 5k条精选通用领域含思维链数据(https://huggingface.co/datasets/Congliu/Chinese-DeepSeek-R1-Distill-data-110k-SFT)+3k条含思维链心理辅导对话(https://huggingface.co/datasets/CAS-SIAT-XinHai/CPsyCoun) | |
| 1. 110k的数据怎么挑选10k条?:基于score | |
| 2. 3k条心理辅导对话:原始对话(alpaca,含history字段,不含思维链,来自于CPsyCounD),使用deepseek r1【通过设置max_token实现低成本获取思维链】给output增加思维链,获得含思维链的数据集 | |
| ### 训练方式 | |
| qlora,变长注意力,数据拼接 | |
| ## 下载模型 | |
| ```bash | |
| git lfs install | |
| git clone https://huggingface.co/juneup/internlm2.5_7b_distill | |
| ``` | |
| 若不想克隆大型文件: | |
| ```bash | |
| GIT_LFS_SKIP_SMUDGE=1 git clone https://huggingface.co/juneup/internlm2.5_7b_distill | |
| ``` | |
| ### 在Ollama下载 | |
| ```bash | |
| ollama run Juneup/internlm2.5_7b_distill:q4_k_m | |