Spaces:
Paused
Paused
| # import gradio as gr | |
| # model_name = "models/THUDM/chatglm2-6b-int4" | |
| # gr.load(model_name).lauch() | |
| # %%writefile demo-4bit.py | |
| from textwrap import dedent | |
| # credit to https://github.com/THUDM/ChatGLM2-6B/blob/main/web_demo.py | |
| # while mistakes are mine | |
| from transformers import AutoModel, AutoTokenizer | |
| import gradio as gr | |
| import mdtex2html | |
| from loguru import logger | |
| model_name = "THUDM/chatglm2-6b" | |
| model_name = "THUDM/chatglm2-6b-int4" | |
| tokenizer = AutoTokenizer.from_pretrained(model_name, trust_remote_code=True) | |
| # model = AutoModel.from_pretrained(model_name, trust_remote_code=True).cuda() | |
| # 4/8 bit | |
| # model = AutoModel.from_pretrained("THUDM/chatglm2-6b", trust_remote_code=True).quantize(4).cuda() | |
| import torch | |
| has_cuda = torch.cuda.is_available() | |
| # has_cuda = False # force cpu | |
| if has_cuda: | |
| model = AutoModel.from_pretrained(model_name, trust_remote_code=True).cuda() # 3.92G | |
| else: | |
| model = AutoModel.from_pretrained(model_name, trust_remote_code=True).half() # .float() .half().float() | |
| model = model.eval() | |
| _ = """Override Chatbot.postprocess""" | |
| def postprocess(self, y): | |
| if y is None: | |
| return [] | |
| for i, (message, response) in enumerate(y): | |
| y[i] = ( | |
| None if message is None else mdtex2html.convert((message)), | |
| None if response is None else mdtex2html.convert(response), | |
| ) | |
| return y | |
| gr.Chatbot.postprocess = postprocess | |
| def parse_text(text): | |
| """copy from https://github.com/GaiZhenbiao/ChuanhuChatGPT/""" | |
| lines = text.split("\n") | |
| lines = [line for line in lines if line != ""] | |
| count = 0 | |
| for i, line in enumerate(lines): | |
| if "```" in line: | |
| count += 1 | |
| items = line.split('`') | |
| if count % 2 == 1: | |
| lines[i] = f'<pre><code class="language-{items[-1]}">' | |
| else: | |
| lines[i] = f'<br></code></pre>' | |
| else: | |
| if i > 0: | |
| if count % 2 == 1: | |
| line = line.replace("`", "\`") | |
| line = line.replace("<", "<") | |
| line = line.replace(">", ">") | |
| line = line.replace(" ", " ") | |
| line = line.replace("*", "*") | |
| line = line.replace("_", "_") | |
| line = line.replace("-", "-") | |
| line = line.replace(".", ".") | |
| line = line.replace("!", "!") | |
| line = line.replace("(", "(") | |
| line = line.replace(")", ")") | |
| line = line.replace("$", "$") | |
| lines[i] = "<br>"+line | |
| text = "".join(lines) | |
| return text | |
| def predict(RETRY_FLAG, input, chatbot, max_length, top_p, temperature, history, past_key_values): | |
| print(f"Inside predict, user_input1 is - {input}") | |
| chatbot.append((parse_text(input), "")) | |
| print(f"Inside predict, user_input2 is - {input}") | |
| print(f"Inside predict, chatbot is - {chatbot}") | |
| print(f"Inside predict, history is - {history}") | |
| #if RETRY_FLAG: | |
| # history.append() | |
| for response, history, past_key_values in model.stream_chat(tokenizer, input, history, past_key_values=past_key_values, | |
| return_past_key_values=True, | |
| max_length=max_length, top_p=top_p, | |
| temperature=temperature): | |
| chatbot[-1] = (parse_text(input), parse_text(response)) | |
| #print(f"Inside predict, chatbot2 is - {chatbot}") | |
| #print(f"Inside predict, history2 is - {history}") | |
| yield chatbot, history, past_key_values | |
| def trans_api(input, max_length=4096, top_p=0.8, temperature=0.2): | |
| if max_length < 100: | |
| max_length = 4096 | |
| if top_p < 0.1: | |
| top_p = 0.8 | |
| if temperature <= 0: | |
| temperature = 0.01 | |
| try: | |
| res, _ = model.chat( | |
| tokenizer, | |
| input, | |
| history=[], | |
| past_key_values=None, | |
| max_length=max_length, | |
| top_p=top_p, | |
| temperature=temperature, | |
| ) | |
| # logger.debug(f"{res=} \n{_=}") | |
| except Exception as exc: | |
| logger.error(f"{exc=}") | |
| res = str(exc) | |
| return res | |
| def reset_user_input(): | |
| return gr.update(value='') | |
| def reset_state(): | |
| return [], [], None | |
| # Delete last turn | |
| def delete_last_turn(chat, history): | |
| if chat and history: | |
| chat.pop(-1) | |
| history.pop(-1) | |
| #history.pop(-1) | |
| return chat, history | |
| # Regenerate response | |
| def retry_last_answer( | |
| user_input, | |
| chatbot, | |
| max_length, | |
| top_p, | |
| temperature, | |
| history, | |
| past_key_values | |
| ): | |
| print(f"chatbot is - {chatbot}") | |
| print(f"history is - {history}") | |
| print(f"user_input is - {user_input}") | |
| print(f"max_length is - {max_length}") | |
| print(f"top_p is - {top_p}") | |
| print(f"temperature is - {temperature}") | |
| print(f"past_key_values type is - {type(past_key_values)}") | |
| if chatbot and history: | |
| # Removing the previous conversation from chat | |
| chatbot.pop(-1) | |
| #chatbot[-1] = (chatbot[-1][0],) | |
| #history[-1] = (history[-1][0],) | |
| # Setting up a flag to capture a retry | |
| RETRY_FLAG = True | |
| # Getting last message from user | |
| user_input = history[-1][0] | |
| # Removing bot response from the history | |
| history.pop(-1) | |
| print(f"popped chatbot is - {chatbot}") | |
| print(f"popped history is - {history}") | |
| #print(f"user_input is - {user_input}") | |
| yield from predict( | |
| RETRY_FLAG, | |
| user_input, | |
| chatbot, | |
| max_length, | |
| top_p, | |
| temperature, | |
| history, | |
| past_key_values | |
| ) | |
| with gr.Blocks(theme=gr.themes.Soft()) as demo: | |
| gr.HTML("""<h1 align="center">ChatGLM2-6B-int4</h1>""") | |
| gr.HTML("""<center><a href="https://huggingface.co/spaces/ysharma/chatglm2-6b-4bit?duplicate=true"><img src="https://bit.ly/3gLdBN6" alt="Duplicate Space"></a>To avoid the queue and for faster inference Duplicate this Space and upgrade to GPU</center>""") | |
| with gr.Accordion("Info", open=False): | |
| _ = """ | |
| A query takes from 30 seconds to a few tens of seconds, dependent on the number of words/characters | |
| the question and answer contain. | |
| * Low temperature: responses will be more deterministic and focused; High temperature: responses more creative. | |
| * Suggested temperatures -- translation: up to 0.3; chatting: > 0.4 | |
| * Top P controls dynamic vocabulary selection based on context. | |
| For a table of example values for different scenarios, refer to [this](https://community.openai.com/t/cheat-sheet-mastering-temperature-and-top-p-in-chatgpt-api-a-few-tips-and-tricks-on-controlling-the-creativity-deterministic-output-of-prompt-responses/172683) | |
| If the instance is not on a GPU (T4), it will be very slow. You can try to run the colab notebook [chatglm2-6b-4bit colab notebook](https://colab.research.google.com/drive/1WkF7kOjVCcBBatDHjaGkuJHnPdMWNtbW?usp=sharing) for a spin. | |
| The T4 GPU is sponsored by a community GPU grant from Huggingface. Thanks a lot! | |
| """ | |
| gr.Markdown(dedent(_)) | |
| chatbot = gr.Chatbot() | |
| with gr.Row(): | |
| with gr.Column(scale=4): | |
| with gr.Column(scale=12): | |
| user_input = gr.Textbox(show_label=False, placeholder="Input...", lines=10).style( | |
| container=False) | |
| RETRY_FLAG = gr.Checkbox(value=False, visible=False) | |
| with gr.Column(min_width=32, scale=1): | |
| with gr.Row(): | |
| submitBtn = gr.Button("Submit", variant="primary") | |
| deleteBtn = gr.Button("Delete last turn", variant="secondary") | |
| retryBtn = gr.Button("Regenerate", variant="secondary") | |
| with gr.Column(scale=1): | |
| emptyBtn = gr.Button("Clear History") | |
| max_length = gr.Slider(0, 32768, value=8192/2, step=1.0, label="Maximum length", interactive=True) | |
| top_p = gr.Slider(0, 1, value=0.8, step=0.01, label="Top P", interactive=True) | |
| temperature = gr.Slider(0.01, 1, value=0.95, step=0.01, label="Temperature", interactive=True) | |
| history = gr.State([]) | |
| past_key_values = gr.State(None) | |
| user_input.submit(predict, [RETRY_FLAG, user_input, chatbot, max_length, top_p, temperature, history, past_key_values], | |
| [chatbot, history, past_key_values], show_progress=True) | |
| submitBtn.click(predict, [RETRY_FLAG, user_input, chatbot, max_length, top_p, temperature, history, past_key_values], | |
| [chatbot, history, past_key_values], show_progress=True, api_name="predict") | |
| submitBtn.click(reset_user_input, [], [user_input]) | |
| emptyBtn.click(reset_state, outputs=[chatbot, history, past_key_values], show_progress=True) | |
| retryBtn.click( | |
| retry_last_answer, | |
| inputs = [user_input, chatbot, max_length, top_p, temperature, history, past_key_values], | |
| #outputs = [chatbot, history, last_user_message, user_message] | |
| outputs=[chatbot, history, past_key_values] | |
| ) | |
| deleteBtn.click(delete_last_turn, [chatbot, history], [chatbot, history]) | |
| with gr.Accordion("For Translation API", open=False): | |
| input_text = gr.Text() | |
| tr_btn = gr.Button("Go", variant="primary") | |
| out_text = gr.Text() | |
| tr_btn.click(trans_api, [input_text, max_length, top_p, temperature], out_text, show_progress=True, api_name="tr") | |
| input_text.submit(trans_api, [input_text, max_length, top_p, temperature], out_text, show_progress=True, api_name="tr") | |
| with gr.Accordion("Example inputs", open=True): | |
| examples = gr.Examples( | |
| examples=[["Explain the plot of Cinderella in a sentence."], | |
| ["How long does it take to become proficient in French, and what are the best methods for retaining information?"], | |
| ["What are some common mistakes to avoid when writing code?"], | |
| ["Build a prompt to generate a beautiful portrait of a horse"], | |
| ["Suggest four metaphors to describe the benefits of AI"], | |
| ["Write a pop song about leaving home for the sandy beaches."], | |
| ["Write a summary demonstrating my ability to do beat-boxing"]], | |
| inputs = [user_input], | |
| ) | |
| # demo.queue().launch(share=False, inbrowser=True) | |
| # demo.queue().launch(share=True, inbrowser=True, debug=True) | |
| demo.queue().launch(debug=True) |