Update app.py

app.py

@@ -18,6 +18,14 @@ import io
 import hashlib
 import datetime
 from utils import build_logger
+from transformers import AutoTokenizer, MistralForCausalLM
+import torch
+import random
+from textwrap import wrap
+from transformers import AutoConfig, AutoTokenizer, AutoModelForSeq2SeqLM, AutoModelForCausalLM, MistralForCausalLM
+from peft import PeftModel, PeftConfig
+import torch
+import os
 
 dotenv.load_dotenv()
 
@@ -312,33 +320,76 @@ def convert_to_markdown(vectara_response_json):
         return f"{markdown_summary}**Sources:**\n{markdown_sources}"
     else:
         return "No data found in the response."
-# Main function to handle the Gradio interface logic
 
-def process_summary_with_openai(summary):
-    """
-    This function takes a summary text as input and processes it with OpenAI's GPT model.
-    """
-    try:
-        # Ensure that the OpenAI client is properly initialized
-        client = OpenAI(api_key=os.getenv('OPENAI_API_KEY'))
-        
-        # Create the prompt for OpenAI's completion
-        prompt = "You are clinical education consultant ready to do just about anything for his students. You are discussing training cases with students at TonicUniversity. You will recieve keywords or an image description. Assess and describe the proper options to your students in minute detail. Propose a course of action for them to base their recommendations on based on your description. You will recieve a summary assessment in a language, always provide a complete answer. Exclude any other commentary:"
-        
-        # Call the OpenAI API with the prompt and the summary
-        completion = client.chat.completions.create(
-            model="gpt-4-1106-preview",  # Make sure to use the correct model name
-            messages=[
-                {"role": "system", "content": prompt},
-                {"role": "user", "content": summary}
-            ]
-        )
-        
-        # Extract the content from the completion
-        final_summary = completion.choices[0].message.content
-        return final_summary
-    except Exception as e:
-        return str(e)
+# Functions to Wrap the Prompt Correctly
+def wrap_text(text, width=90):
+    lines = text.split('\n')
+    wrapped_lines = [textwrap.fill(line, width=width) for line in lines]
+    wrapped_text = '\n'.join(wrapped_lines)
+    return wrapped_text
+def multimodal_prompt(user_input, system_prompt="You are an expert medical analyst:"):
+
+    # Combine user input and system prompt
+    formatted_input = f"{user_input}{system_prompt}"
+
+    # Encode the input text
+    encodeds = tokenizer(formatted_input, return_tensors="pt", add_special_tokens=False)
+    model_inputs = encodeds.to(device)
+
+    # Generate a response using the model
+    output = model.generate(
+        **model_inputs,
+        max_length=max_length,
+        use_cache=True,
+        early_stopping=True,
+        bos_token_id=model.config.bos_token_id,
+        eos_token_id=model.config.eos_token_id,
+        pad_token_id=model.config.eos_token_id,
+        temperature=0.1,
+        do_sample=True
+    )
+
+    # Decode the response
+    response_text = tokenizer.decode(output[0], skip_special_tokens=True)
+
+    return response_text
+
+# Define the device
+device = "cuda" if torch.cuda.is_available() else "cpu"
+
+# Use the base model's ID
+base_model_id = "stabilityai/stablelm-3b-4e1t"
+model_directory = "Tonic/stablemed"
+
+# Instantiate the Tokenizer
+tokenizer = AutoTokenizer.from_pretrained("stabilityai/stablelm-3b-4e1t", token=hf_token, trust_remote_code=True, padding_side="left")
+# tokenizer = AutoTokenizer.from_pretrained("Tonic/stablemed", trust_remote_code=True, padding_side="left")
+tokenizer.pad_token = tokenizer.eos_token
+tokenizer.padding_side = 'left'
+
+# Load the PEFT model
+peft_config = PeftConfig.from_pretrained("Tonic/stablemed", token=hf_token)
+peft_model = AutoModelForCausalLM.from_pretrained("stabilityai/stablelm-3b-4e1t", token=hf_token, trust_remote_code=True)
+peft_model = PeftModel.from_pretrained(peft_model, "Tonic/stablemed", token=hf_token)
+
+
+class ChatBot:
+    def __init__(self):
+        self.history = []
+
+    def predict(self, user_input, system_prompt="You are an expert medical analyst:"):
+        formatted_input = f"<s>[INST]{system_prompt} {user_input}[/INST]"
+        user_input_ids = tokenizer.encode(formatted_input, return_tensors="pt")
+        response = peft_model.generate(input_ids=user_input_ids, max_length=512, pad_token_id=tokenizer.eos_token_id)
+        response_text = tokenizer.decode(response[0], skip_special_tokens=True)
+        return response_text
+
+def process_summary_with_stablemed(summary):
+    system_prompt = "You are a clinical education consultant ready to do just about anything for your students. You are discussing training cases with students at TonicUniversity. Assess and describe the proper options to your students in minute detail. Propose a course of action for them to base their recommendations on based on your description. Always provide a complete answer. Exclude any other commentary:"
+    response_text = bot.predict(summary, system_prompt)
+    return response_text
+
+# Main function to handle the Gradio interface logic
 
 def process_and_query(input_language=None, audio_input=None, image_input=None, text_input=None):
     try:
@@ -378,7 +429,7 @@ def process_and_query(input_language=None, audio_input=None, image_input=None, t
             markdown_output += "\n\n**Original Image Description:**\n" + image_description
         
         # Process the summary with OpenAI
-        final_response = process_summary_with_openai(markdown_output)
+        final_response = process_summary_with_stablemed(markdown_output)
         print("Final Response:", final_response)  # Debug print
 
         # Evaluate hallucination
@@ -396,15 +447,17 @@ def process_and_query(input_language=None, audio_input=None, image_input=None, t
 welcome_message = """
 # 👋🏻Welcome to ⚕🗣️😷MultiMed - Access Chat ⚕🗣️😷
 
-🗣️📝 This is an educational and accessible conversational tool to improve wellness and sanitation in support of public health. 
+🗣️📝 This is an educational and accessible conversational tool.
 
 ### How To Use ⚕🗣️😷MultiMed⚕: 
 
 🗣️📝Interact with ⚕🗣️😷MultiMed⚕ in any language using image, audio or text!
 
-📚🌟💼 The quality of the answers depends on the quality of the dataset, so if you want to see some data represented here, do [get in touch](https://discord.gg/GWpVpekp). You can also use 😷MultiMed⚕️ on your own data & in your own way by cloning this space. 🧬🔬🔍 Simply click here: <a style="display:inline-block" href="https://huggingface.co/spaces/TeamTonic/MultiMed?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>
+📚🌟💼 that uses [Tonic/stablemed](https://huggingface.co/Tonic/stablemed) and [adept/fuyu-8B](https://huggingface.co/adept/fuyu-8b) with [Vectara](https://huggingface.co/vectara) embeddings + retrieval. 
+do [get in touch](https://discord.gg/GWpVpekp). You can also use 😷MultiMed⚕️ on your own data & in your own way by cloning this space. 🧬🔬🔍 Simply click here: <a style="display:inline-block" href="https://huggingface.co/spaces/TeamTonic/MultiMed?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 : 
 
-#### Join us : 🌟TeamTonic🌟 is always making cool demos! Join our active builder's🛠️community on 👻Discord: [Discord](https://discord.gg/GWpVpekp) On 🤗Huggingface: [TeamTonic](https://huggingface.co/TeamTonic) & [MultiTransformer](https://huggingface.co/MultiTransformer) On 🌐Github: [Polytonic](https://github.com/tonic-ai) & contribute to 🌟 [PolyGPT](https://github.com/tonic-ai/polygpt-alpha)"             
+🌟TeamTonic🌟 is always making cool demos! Join our active builder's🛠️community on 👻Discord: [Discord](https://discord.gg/GWpVpekp) On 🤗Huggingface: [TeamTonic](https://huggingface.co/TeamTonic) & [MultiTransformer](https://huggingface.co/MultiTransformer) On 🌐Github: [Polytonic](https://github.com/tonic-ai) & contribute to 🌟 [PolyGPT](https://github.com/tonic-ai/polygpt-alpha)"             
 """
 
 
@@ -520,14 +573,10 @@ with gr.Blocks(theme='ParityError/Anime') as iface :
         input_language = gr.Dropdown(languages, label="select the language",value="English",interactive=True)
         audio_input = gr.Audio(label="speak",type="filepath",sources="microphone")
         audio_output = gr.Markdown(label="output text")
-        # audio_button = gr.Button("process audio")
-        # audio_button.click(process_speech, inputs=[input_language,audio_input], outputs=audio_output)
         gr.Examples([["English","sample_input.mp3"]],inputs=[input_language,audio_input])
     with gr.Accordion("image identification",open=True):
         image_input = gr.Image(label="upload image")
         image_output = gr.Markdown(label="output text")
-        # image_button = gr.Button("process image")
-        # image_button.click(process_image, inputs=image_input, outputs=image_output)
         gr.Examples(["sick person.jpeg"],inputs=[image_input])
     with gr.Accordion("text summarization",open=True):
         text_input = gr.Textbox(label="input text",lines=5)
@@ -545,15 +594,5 @@ with gr.Blocks(theme='ParityError/Anime') as iface :
             ["구강 헤르페스의 적절한 치료법은 무엇입니까?"],
             ["Je, ni matibabu gani sahihi kwa herpes ya buccal?"],
         ],inputs=[text_input])
-    # with gr.Accordion("hallucination check",open=True):
-    #     assertion = gr.Textbox(label="assertion")
-    #     citation =  gr.Textbox(label="citation text")
-    #     hullucination_output = gr.Markdown(label="output text")
-    #     hallucination_button = gr.Button("check hallucination")
-    #     gr.Examples([["i am drunk","sarah is pregnant"]],inputs=[assertion,citation])
-    #     hallucination_button.click(check_hallucination,inputs=[assertion,citation],outputs=hullucination_output)
-    
-
-
 
 iface.queue().launch(show_error=True,debug=True)