File size: 5,930 Bytes
7e59293 71f3ad4 7e59293 d0c4824 7e59293 d0c4824 7e59293 d0c4824 7e59293 d27b26a 7e59293 d27b26a 7e59293 14ddd0a 7e59293 d0c4824 7e59293 f7aa7fc d0c4824 f7aa7fc d0c4824 f7aa7fc d0c4824 7e59293 d0c4824 7e59293 71f3ad4 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 |
---
pipeline_tag: text-generation
---
# Collective roleplay model
<!-- Provide a quick summary of what the model is/does. -->
Model developed by Collective AI for role-playing with learning from user interactions data.
### Model Description
<!-- Provide a longer summary of what this model is. -->
- **Developed by:** [Collective AI](https://huggingface.co/collective-ai)
- **Model type:** llama3 based role-play model
- **Language(s):** Chinese, English
- **Finetuned from model:** [Llama3-8B-Chinese-Chat](https://huggingface.co/shenzhi-wang/Llama3-8B-Chinese-Chat)
## How to Get Started with the Model
Requirements
```
transformers>=4.40.2
```
Use the code below to get started with the model.
```python
from transformers import AutoTokenizer, AutoModelForCausalLM
model_id = "Collective-Ai/collective-v0.1-chinese-roleplay-8b"
tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(
model_id, torch_dtype="auto", device_map="auto"
)
name = "唐三"
gendar = "Male"
age = "19"
personality = "孤傲，冷漠，沉着，冷静"
relation_between_role_and_user = "同门师兄弟"
role_base_story = "《斗罗大陆》男主角。前世为唐门外门弟子，因偷学内门绝学《玄天宝录》，为唐门所不容，跳崖明志，却来到了另一个世界——斗罗大陆"
messages = [
{"role": "system", "content": f"""#Role\nName: {name}\nGender: {gendar}\nLanguage: Chinese\nAge: {age}\nPersonality: {personality}\n\n#Relationship\n{relation_between_role_and_user}\n\n#Story\n{role_base_story}"""},
{"role": "user", "content": "你好"},
]
input_ids = tokenizer.apply_chat_template(messages, add_generation_prompt=True, return_tensors="pt").to(model.device)
outputs = model.generate(
input_ids,
max_new_tokens=8192,
do_sample=True,
temperature=1.0,
top_p=0.7,
)
response = outputs[0][input_ids.shape[-1]:]
print(tokenizer.decode(response, skip_special_tokens=True))
```
<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
## Prompt
### Huggingface inference API
<!-- This section is used for prompt online -->
To get better model experience, using the template below to assemble your prompt:
```python
def format_message(role, message):
return f"<|start_header_id|>{role}<|end_header_id|>\n\n{message}<|eot_id|>"
# history: 历史对话
# instruction: 角色人物卡
item = {'history':[{'role':'user','content':'你好'},{'role':'assistant','content':'你好,我是唐三'},{'role':'user','content':'带我去学院吧'}],
'instruction':'#Role\nName: 唐三\nGender: Male\nLanguage: Chinese\nAge: 19\nPersonality: 孤傲，冷漠，沉着，冷静\n\n#Relationship\n同门师兄弟\n\n#Story\n《斗罗大陆》男主角。前世为唐门外门弟子，因偷学内门绝学《玄天宝录》，为唐门所不容，跳崖明志，却来到了另一个世界——斗罗大陆'}
histories = item.get('history', [])
if histories == [] or histories == [{}]:
instruction = ''
else:
instruction = ''.join(format_message(hist['role'], hist['content']) for hist in histories)
system = '<|begin_of_text|><|start_header_id|>system<|end_header_id|>\n'+item.get('instruction', '') + '<|eot_id|>' + instruction + '<|start_header_id|>assistant<|end_header_id|>'
```
The final prompt should look like this:(Since this is a role-playing model, your prompt should better includ role info and story)
```
<|begin_of_text|><|start_header_id|>system<|end_header_id|>
#Role
Name: 唐三
Gender: Male
Language: Chinese
Age: 19
Personality: 孤傲，冷漠，沉着，冷静
#Relationship
同门师兄弟
#Story
《斗罗大陆》男主角。前世为唐门外门弟子，因偷学内门绝学《玄天宝录》，为唐门所不容，跳崖明志，却来到了另一个世界——斗罗大陆<|eot_id|><|start_header_id|>user<|end_header_id|>
你好<|eot_id|><|start_header_id|>assistant<|end_header_id|>
你好,我是唐三<|eot_id|><|start_header_id|>user<|end_header_id|>
带我去学院吧<|eot_id|><|start_header_id|>assistant<|end_header_id|>
```
### Official API
<!-- This section is used for official api -->
Coming soon
## Improvement
Our model demonstrates around 150% improvement over ChatGPT in terms of average conversation length. It can extend the average dialog turns from 80 replies to as many as 200 turns or more, significantly improving user engagement.
Our chatbot model has several distinctive strengths:
**Rich Expressiveness**: The model is capable of conveying semantic and emotional nuances by utilizing detailed descriptions within the dialogue, such as speech expression, non-verbal cues, and character psychological portrayals, complementing the insufficiency of pure textual information.
**Proactivity**: The model can not only flexibly engage with users’ diverse inputs, but also proactively introduce new topics, which greatly improved user retention and depth of conversations. It's also less susceptible to repetitions.
**Strong Linguistic Foundations**: By leveraging massive high-quality Chinese corpora, the model can engage in diverse conversations based on different scenarios and character attributes, including genres like historical, campus, workplace, and fantasy, providing users with richer chat experiences.
Here are some example roleplay dialogues from our model:
 |