You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Copy file name to clipboardexpand all lines: README_EN.md
+12-6
Original file line number
Diff line number
Diff line change
@@ -69,7 +69,9 @@
69
69
70
70
</div>
71
71
72
-
Everyone is welcome to contribute to this project ~
72
+
🎉 Everyone is welcome to contribute to this project!
73
+
74
+
🔍 Those who are interested in the principles and underlying implementations of LLMs can follow [ThinkLLM](https://github.com/aJupyter/ThinkLLM), which focuses on building various components of large models from scratch.
73
75
74
76
---
75
77
@@ -105,6 +107,8 @@ The Model aims to fully understand and promote the mental health of individuals,
105
107
</table>
106
108
107
109
## Recent Updates
110
+
-[2025.3] Based on the full fine-tuning of InternLM2.5-7B-chat, [EmoLLM (in GGUF format, fp16 precision)](https://huggingface.co/collections/L0ve1ace/psychology-llm-gguf-67cc766eaf0a3f01c6e39aa6) has been released. Updates on how to operate it will be provided later. @Rycen7822@Slipstream-Max
111
+
-[2025.2] Updated the [first mental health R1 distillation dataset](./datasets/psychology-10k-Deepseek-R1-zh.json)@Kedreamix
108
112
-[2024.09.14] The Lora fine-tuned model based on the Qwen2-7B-Instruct model is open-sourced. Fine-tuning configuration file address: [Qwen2-7B-Instruct_lora.py](./xtuner_config/Qwen2-7B-Instruct_lora.py), model weight link: [ModelScope](https://www.modelscope.cn/models/aJupyter/EmoLLM_Qwen2-7B-Instruct_lora/)
109
113
-[2024.08] The Lora fine-tuned model based on GLM4-9B-chat is open-sourced (based on Llama-factory). For details, see [Fine-tuning Tutorial](./doc/GLM-4-9B-chat%20Lora%20微调(llama-factory).md), model weight link: [ModelScope](https://www.modelscope.cn/models/wwewwt/EmoLLM-glm-4-9b-chat/summary)
110
114
-[2024.07.16] Welcome everyone to experience EmoLLM V3.0. This model is a fully fine-tuned version based on the InternLM2.5-7B-Chat model. The fine-tuning configuration file can be found at: [internlm2_5_chat_7b_full.py](./xtuner_config/internlm2_5_chat_7b_full.py). Model weights are available at: [OpenXLab](https://openxlab.org.cn/models/detail/chg0901/EmoLLM_V3.0), [ModelScope](https://modelscope.cn/models/chg0901/EmoLLMV3.0). WebDemo is available at: [OpenXLab apps](https://openxlab.org.cn/apps/detail/chg0901/EmoLLMV3.0), [Full fine-tuning tutorial on Zhihu](https://zhuanlan.zhihu.com/p/708931911).
@@ -120,16 +124,18 @@ The Model aims to fully understand and promote the mental health of individuals,
120
124
-[2024.04.02] Uploaded at Huggingface [Old Mother Counsellor](https://huggingface.co/brycewang2018/EmoLLM-mother/tree/main)
121
125
-[2024.03.25][Mother-like Therapist] is released on Huggingface (https://huggingface.co/brycewang2018/EmoLLM-mother/tree/main)
122
126
-[2024.03.25][Daddy-like Boy-Friend] is released on Baidu Paddle-Paddle AI Studio Platform (https://aistudio.baidu.com/community/app/68787)
123
-
-[2024.03.24] The **InternLM2-Base-7B QLoRA fine-tuned model** has been released on the **OpenXLab** and **ModelScope** platforms. For more details, please refer to [**InternLM2-Base-7B QLoRA**](./xtuner_config/README_internlm2_7b_base_qlora.md).
124
-
-[2024.03.12][aiwei] is released on Baidu Paddle-Paddle AI Studio Platform (https://aistudio.baidu.com/community/app/63335)
125
-
-[2024.03.11]**EmoLLM V2.0 is greatly improved in all scores compared to EmoLLM V1.0. Surpasses the performance of Role-playing ChatGPT on counseling tasks!**[Click to experience EmoLLM V2.0](https://openxlab.org.cn/apps/detail/Farewell1/EmoLLMV2.0), update [dataset statistics and details](./datasets/), [Roadmap](./assets/Roadmap_ZH.png)
-[2024.03.03][Based on InternLM2-7B-chat full fine-tuned version EmoLLM V2.0 open sourced](https://openxlab.org.cn/models/detail/ajupyter/EmoLLM_internlm2_7b_full), need two A100*80G, update professional evaluation, see [evaluate](./evaluate/), update PaddleOCR-based PDF to txt tool scripts, see [scripts](./scripts/).
127
+
128
128
129
129
130
130
<details>
131
131
<summary>View More</summary>
132
132
133
+
134
+
-[2024.03.24] The **InternLM2-Base-7B QLoRA fine-tuned model** has been released on the **OpenXLab** and **ModelScope** platforms. For more details, please refer to [**InternLM2-Base-7B QLoRA**](./xtuner_config/README_internlm2_7b_base_qlora.md).
135
+
-[2024.03.12][aiwei] is released on Baidu Paddle-Paddle AI Studio Platform (https://aistudio.baidu.com/community/app/63335)
136
+
-[2024.03.11]**EmoLLM V2.0 is greatly improved in all scores compared to EmoLLM V1.0. Surpasses the performance of Role-playing ChatGPT on counseling tasks!**[Click to experience EmoLLM V2.0](https://openxlab.org.cn/apps/detail/Farewell1/EmoLLMV2.0), update [dataset statistics and details](./datasets/), [Roadmap](./assets/Roadmap_ZH.png)
-[2024.03.03][Based on InternLM2-7B-chat full fine-tuned version EmoLLM V2.0 open sourced](https://openxlab.org.cn/models/detail/ajupyter/EmoLLM_internlm2_7b_full), need two A100*80G, update professional evaluation, see [evaluate](./evaluate/), update PaddleOCR-based PDF to txt tool scripts, see [scripts](./scripts/).
133
139
-[2024.02.29] Updated objective assessment calculations, see [evaluate](./evaluate/) for details. A series of datasets have also been updated, see [datasets](./datasets/) for details.
134
140
-[2024.02.27] Updated English README and a series of datasets (licking dogs and one-round dialogue)
135
141
-[2024.02.23]The "Gentle Lady Psychologist Ai Wei" based on InternLM2_7B_chat_qlora was launched. [Click here to obtain the model weights](https://openxlab.org.cn/models/detail/ajupyter/EmoLLM_aiwei), [configuration file](xtuner_config/aiwei-internlm2_chat_7b_qlora.py), [online experience link](https://openxlab.org.cn/apps/detail/ajupyter/EmoLLM-aiwei)
- 【2024.6】[LLaMA-Factory](https://github.com/hiyouga/LLaMA-Factory)の[GLM4-9B-chat微調整ガイド](./doc/GLM-4-9B-chat%20Lora%20微调(llama-factory).md)を追加しました。[swiftベースの微調整ガイド](./swift/)を追加しました。論文[ESC-Eval: Evaluating Emotion Support Conversations in Large Language Models](https://arxiv.org/abs/2406.14952)がEmoLLMを引用し、EmoLLMが良好な結果を達成しました。
0 commit comments