| import gradio as gr | |
| from gpt4all import GPT4All | |
| from huggingface_hub import hf_hub_download | |
| title = "安风" | |
| description = """ | |
| Infer service | |
| """ | |
| model_path = "tastypear/CausalLM-7B-DPO-alpha-GGUF" | |
| model_name = "causallm_7b-dpo-alpha.Q4_K_M.gguf" | |
| hf_hub_download(repo_id="tastypear/CausalLM-7B-DPO-alpha-GGUF", filename=model_name, local_dir=model_path, local_dir_use_symlinks=True) | |
| print("Start the model init process") | |
| model = model = GPT4All(model_name, model_path, allow_download = True, device="cpu") | |
| print("Finish the model init process") | |
| model.config["promptTemplate"] = '''<|im_start|>user | |
| {0}<|im_end|> | |
| <|im_start|>assistant | |
| ''' | |
| model.config["systemPrompt"] = "You are a helpful assistant named 安风." | |
| model._is_chat_session_activated = True | |
| max_new_tokens = 2048 | |
| def generater(message, history, temperature, top_p, top_k): | |
| prompt = "" | |
| for user_message, assistant_message in history: | |
| prompt += model.config["promptTemplate"].format(user_message) | |
| prompt += assistant_message + "<|im_end|>" | |
| prompt += model.config["promptTemplate"].format(message) | |
| outputs = [] | |
| for token in model.generate(prompt=prompt, temp=temperature, top_k = top_k, top_p = top_p, max_tokens = max_new_tokens, streaming=True): | |
| outputs.append(token) | |
| yield "".join(outputs) | |
| def vote(data: gr.LikeData): | |
| if data.liked: | |
| return | |
| else: | |
| return | |
| chatbot = gr.Chatbot(avatar_images=('resourse/user-icon.png', 'resourse/chatbot-icon.png'),bubble_full_width = False) | |
| additional_inputs=[ | |
| gr.Slider( | |
| label="temperature", | |
| value=0.5, | |
| minimum=0.0, | |
| maximum=2.0, | |
| step=0.05, | |
| interactive=True, | |
| info="Higher values like 0.8 will make the output more random, while lower values like 0.2 will make it more focused and deterministic.", | |
| ), | |
| gr.Slider( | |
| label="top_p", | |
| value=1.0, | |
| minimum=0.0, | |
| maximum=1.0, | |
| step=0.01, | |
| interactive=True, | |
| info="0.1 means only the tokens comprising the top 10% probability mass are considered. Suggest set to 1 and use temperature. 1 means 100% and will disable it", | |
| ), | |
| gr.Slider( | |
| label="top_k", | |
| value=40, | |
| minimum=0, | |
| maximum=1000, | |
| step=1, | |
| interactive=True, | |
| info="limits candidate tokens to a fixed number after sorting by probability. Setting it higher than the vocabulary size deactivates this limit.", | |
| ) | |
| ] | |
| iface = gr.ChatInterface( | |
| fn = generater, | |
| title=title, | |
| description = description, | |
| additional_inputs=additional_inputs, | |
| examples=[ | |
| ["Can you tell me how the Namib Desert Beetle inspires water collection methods?"], | |
| ["I'm working on a project related to sustainable architecture. How can biomimicry guide my design process?"], | |
| ["Can you explain the concept of biomimicry and its importance in today’s world?"], | |
| ["I need some ideas for a biomimicry project in my biology class. Can you suggest some organisms to study?"], | |
| ["How does the structure of a lotus leaf help in creating self-cleaning surfaces?"] | |
| ] | |
| ) | |
| with gr.Blocks(css="resourse/style/custom.css") as demo: | |
| chatbot.like(vote, None, None) | |
| iface.render() | |
| if __name__ == "__main__": | |
| demo.queue().launch() | |