initial commit

app.py

@@ -1,90 +1,76 @@
 import spaces
-import gradio as gr
 import torch
 import sys
 import html
 from transformers import AutoTokenizer, AutoModelForCausalLM, TextIteratorStreamer
 from threading import Thread
+import gradio as gr
+from gradio_rich_textbox import RichTextbox
 
-model_name_or_path = 'TencentARC/Mistral_Pro_8B_v0.1'
-
-tokenizer = AutoTokenizer.from_pretrained(model_name_or_path, use_fast=False)
-model = AutoModelForCausalLM.from_pretrained(model_name_or_path)
 
-model.half().cuda()
 
-def convert_message(message):
-    message_text = ""
-    if message["content"] is None and message["role"] == "assistant":
-        message_text += "<|assistant|>\n"  # final msg
-    elif message["role"] == "system":
-        message_text += "<|system|>\n" + message["content"].strip() + "\n"
-    elif message["role"] == "user":
-        message_text += "<|user|>\n" + message["content"].strip() + "\n"
-    elif message["role"] == "assistant":
-        message_text += "<|assistant|>\n" + message["content"].strip() + "\n"
-    else:
-        raise ValueError("Invalid role: {}".format(message["role"]))
-    # gradio cleaning - it converts stuff to html entities
-    # we would need special handling for where we want to keep the html...
-    message_text = html.unescape(message_text)
-    # it also converts newlines to <br>, undo this.
-    message_text = message_text.replace("<br>", "\n")
-    return message_text
+title = """# 🙋🏻‍♂️Welcome to🌟Tonic's🫡📉MetaMath
+this is Tencent's mistral DPO finetune for mathematics. You can build with this endpoint using🫡📉MetaMath available here : [TencentARC/Mistral_Pro_8B_v0.1](https://huggingface.co/TencentARC/Mistral_Pro_8B_v0.1). We're using 🤖[introspector/unimath](https://huggingface.co/datasets/introspector/unimath) for cool examples, check it out below ! The demo is still a work in progress and we're looking forward to build downstream tasks that showcase outstanding mathematical reasoning. Have any ideas ? join us below !
+You can also use 🫡📉MetaMath by cloning this space. Simply click here: <a style="display:inline-block" href="https://huggingface.co/spaces/Tonic/MetaMath?duplicate=true"><img src="https://img.shields.io/badge/-Duplicate%20Space-blue?labelColor=white&style=flat&logo=data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAABAAAAAQCAYAAAAf8/9hAAAAAXNSR0IArs4c6QAAAP5JREFUOE+lk7FqAkEURY+ltunEgFXS2sZGIbXfEPdLlnxJyDdYB62sbbUKpLbVNhyYFzbrrA74YJlh9r079973psed0cvUD4A+4HoCjsA85X0Dfn/RBLBgBDxnQPfAEJgBY+A9gALA4tcbamSzS4xq4FOQAJgCDwV2CPKV8tZAJcAjMMkUe1vX+U+SMhfAJEHasQIWmXNN3abzDwHUrgcRGmYcgKe0bxrblHEB4E/pndMazNpSZGcsZdBlYJcEL9Afo75molJyM2FxmPgmgPqlWNLGfwZGG6UiyEvLzHYDmoPkDDiNm9JR9uboiONcBXrpY1qmgs21x1QwyZcpvxt9NS09PlsPAAAAAElFTkSuQmCC&logoWidth=14" alt="Duplicate Space"></a></h3> 
+Join us : 🌟TeamTonic🌟 is always making cool demos! Join our active builder's 🛠️community 👻 [![Join us on Discord](https://img.shields.io/discord/1109943800132010065?label=Discord&logo=discord&style=flat-square)](https://discord.gg/GWpVpekp) On 🤗Huggingface: [TeamTonic](https://huggingface.co/TeamTonic) & [MultiTransformer](https://huggingface.co/MultiTransformer) Math with [introspector](https://huggingface.co/introspector) On 🌐Github: [Tonic-AI](https://github.com/tonic-ai) & contribute to🌟 [SciTonic](https://github.com/Tonic-AI/scitonic)🤗Big thanks to Yuvi Sharma and all the folks at huggingface for the community grant 🤗
+"""
 
-def convert_history(chat_history, max_input_length=1024):
-    history_text = ""
-    idx = len(chat_history) - 1
-    # add messages in reverse order until we hit max_input_length
-    while len(tokenizer(history_text).input_ids) < max_input_length and idx >= 0:
-        user_message, chatbot_message = chat_history[idx]
-        user_message = convert_message({"role": "user", "content": user_message})
-        chatbot_message = convert_message({"role": "assistant", "content": chatbot_message})
-        history_text = user_message + chatbot_message + history_text
-        idx = idx - 1
-    # if nothing was added, add <|assistant|> to start generation.
-    if history_text == "":
-        history_text = "<|assistant|>\n"
-    return history_text
+model_name = 'TencentARC/Mistral_Pro_8B_v0.1'
+tokenizer = AutoTokenizer.from_pretrained(model_name)
+model = AutoModelForCausalLM.from_pretrained(model_name, torch_dtype=torch.bfloat16, device_map="auto")
+model.generation_config = GenerationConfig.from_pretrained(model_name)
+model.generation_config.pad_token_id = model.generation_config.eos_token_id
 
-@spaces.GPU
 @torch.inference_mode()
-def instruct(instruction, max_token_output=1024):
-    input_text = instruction
-    input_ids = tokenizer(input_text, return_tensors='pt', truncation=True).to('cuda')
-    outputs = model.generate(**input_ids, max_length=max_token_output, do_sample=False)
-    return tokenizer.decode(outputs[0], skip_special_tokens=True)
-
+@spaces.GPU
+def predict_math_bot(user_message, system_message, max_new_tokens, temperature, top_p, repetition_penalty):
+    prompt = f"<|system|>\n{self.system_message}\n<|user|>\n{user_message}<|assistant|>" if system_message else user_message
+    inputs = tokenizer(prompt, return_tensors='pt', add_special_tokens=False)
+    input_ids = inputs["input_ids"].to(model.device)
 
-with gr.Blocks() as demo:
-    # chatbot-style model
-    with gr.Tab("Chatbot"):
-        chatbot = gr.Chatbot([], elem_id="chatbot")
-        msg = gr.Textbox()
-        clear = gr.Button("Clear")
-        send_button = gr.Button("Send")  # Add a send button
-        # fn to add user message to history
-        def user(user_message, history):
-            return "", history + [[user_message, None]]
+    output_ids = model.generate(
+        input_ids,
+        max_length=input_ids.shape[1] + max_new_tokens,
+        temperature=temperature,
+        top_p=top_p,
+        repetition_penalty=repetition_penalty,
+        pad_token_id=tokenizer.eos_token_id,
+        do_sample=True
+    )
 
-    def bot(history):
-        prompt = convert_history(history)
-        streaming_out = instruct(prompt)
-        history[-1][1] = ""
-        for new_token in streaming_out:
-            history[-1][1] += new_token
-            yield history
+    response = tokenizer.decode(output_ids[0], skip_special_tokens=True)
+    return response
 
-    msg.submit(user, [msg, chatbot], [msg, chatbot], queue=False).then(
-        bot, chatbot, chatbot
-    )
+def main():
+    with gr.Blocks() as demo:
+        gr.Markdown(title)
+        output_text = gr.RichTextbox(label="🫡📉MetaMath", showlabel=True, interactive=True)
+        
+        with gr.Accordion("Advanced Settings"):
+            with gr.Row():
+                max_new_tokens = gr.Slider(label="Max new tokens", value=125, minimum=25, maximum=1250)
+                temperature = gr.Slider(label="Temperature", value=0.1, minimum=0.05, maximum=1.0)
+                top_p = gr.Slider(label="Top-p (nucleus sampling)", value=0.90, minimum=0.01, maximum=0.99)
+                repetition_penalty = gr.Slider(label="Repetition penalty", value=1.9, minimum=1.0, maximum=2.0)
+                use_custom_settings = gr.Checkbox(label="Use custom settings", value=False)
+        
+        with gr.Row():
+            user_message = gr.Textbox(label="🫡Your Message", lines=3, placeholder="Enter your math query here...")
+            system_message = gr.Textbox(label="📉System Prompt", lines=2, placeholder="Optional: Set a scene or introduce a character...")
 
-    send_button.click(user, [msg, chatbot], [msg, chatbot], queue=False).then(
-            bot, chatbot, chatbot
+        gr.Button("Generate").click(
+            predict_math_bot,
+            inputs=[user_message, system_message, max_new_tokens, temperature, top_p, repetition_penalty, use_custom_settings],
+            outputs=output_text
         )
 
-    clear.click(lambda: None, None, chatbot, queue=False)
+        # Example placeholders
+        demo.add_examples([
+            ["Calculate the integral of x^2 from 0 to 1", "produce a complete mathematical explanation of the following in rich text format:", 125, 0.1, 0.9, 1.9, False],
+            ["What is the derivative of sin(x)?", "produce a complete mathematical explanation of the following in rich text format:", 125, 0.1, 0.9, 1.9, False]
+        ])
 
+    demo.launch()
 
 if __name__ == "__main__":
-    demo.queue().launch(share=True)
+    main()

requirements.txt

@@ -1,8 +1,8 @@
+torch
+transformers
 accelerate
 bitsandbytes
-gradio
 protobuf
 scipy
 sentencepiece
-torch
-git+https://github.com/huggingface/transformers@main
+gradio_rich_textbox