Update app.py

app.py

@@ -1,99 +1,92 @@
+import optimum
 import transformers
-from transformers import AutoTokenizer, MistralForCausalLM
-from transformers import AutoConfig, AutoTokenizer, AutoModelForSeq2SeqLM, AutoModelForCausalLM, MistralForCausalLM
+from transformers import AutoConfig, AutoTokenizer, AutoModel, AutoModelForCausalLM
+from optimum.bettertransformer import BetterTransformer
 import torch
 import gradio as gr
-import random
-from textwrap import wrap
-from peft import PeftModel, PeftConfig
-import torch
-import gradio as gr
-
-# 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"{system_prompt} {user_input}"
+import json
+import os
+import shutil
+import requests
 
-    # 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
-    )
+title = "👋🏻Welcome to Tonic's 💫🌠Starling 7B"
+description = "You can use [💫🌠Starling 7B](https://huggingface.co/berkeley-nest/Starling-RM-7B-alpha) or duplicate it for local use or on Hugging Face! [Join me on Discord to build together](https://discord.gg/VqTxc76K3u)."
+examples = [
+    [
+        "The following dialogue is a conversation between Emmanuel Macron and Elon Musk:",  # user_message
+        "[Emmanuel Macron]: Hello Mr. Musk. Thank you for receiving me today.",  # assistant_message
+        0.9,  # temperature
+        450,  # max_new_tokens
+        0.90,  # top_p
+        1.9,  # repetition_penalty
+    ]
+]
 
-    response_text = tokenizer.decode(output[0], skip_special_tokens=True)
+model_name = "berkeley-nest/Starling-RM-7B-alpha"
 
-    return response_text
 
 device = "cuda" if torch.cuda.is_available() else "cpu"
-model_name = "OpenLLM-France/Claire-Mistral-7B-0.1"
+temperature=0.4
+max_new_tokens=240
+top_p=0.92
+repetition_penalty=1.7
+
 
 tokenizer = transformers.AutoTokenizer.from_pretrained(model_name)
 model = transformers.AutoModelForCausalLM.from_pretrained(model_name,
     device_map="auto",
     torch_dtype=torch.bfloat16,
-    load_in_4bit=True                          # For efficient inference, if supported by the GPU card
+    load_in_4bit=True
 )
-
-class ChatBot:
-    def __init__(self):
-        self.history = []
-
-    def predict(self, user_input, system_prompt):
-        # Combine user input and system prompt
-        formatted_input = f"{system_prompt} {user_input}"
-
-        # Encode user input
-        user_input_ids = tokenizer.encode(formatted_input, return_tensors="pt")
-
-        # Concatenate the user input with chat history
-        if len(self.history) > 0:
-            chat_history_ids = torch.cat([self.history, user_input_ids], dim=-1)
-        else:
-            chat_history_ids = user_input_ids
-
-        # Generate a response using the PEFT model
-        response = model.generate(input_ids=chat_history_ids, max_length=200, pad_token_id=tokenizer.eos_token_id)
-
-        # Update chat history
-        self.history = chat_history_ids
-
-        # Decode and return the response
+model = BetterTransformer.transform(model)
+
+class StarlingBot:
+    def __init__(self, system_prompt="The following dialogue is a conversation"):
+        self.system_prompt = system_prompt
+
+    def predict(self, user_message, assistant_message, system_prompt, do_sample, temperature=0.4, max_new_tokens=700, top_p=0.99, repetition_penalty=1.9):
+        conversation = f" <s> [INST] {self.system_prompt} [INST]  {assistant_message if assistant_message else ''} </s> [/INST]  {user_message}  </s> "
+        # Encode the conversation using the tokenizer
+        input_ids = tokenizer.encode(conversation, return_tensors="pt", add_special_tokens=False)
+        input_ids = input_ids.to(device)
+        response = model.generate(
+            input_ids=input_ids, 
+            use_cache=False, 
+            early_stopping=False, 
+            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=temperature, 
+            do_sample=True, 
+            max_new_tokens=max_new_tokens, 
+            top_p=top_p, 
+            repetition_penalty=repetition_penalty
+        )
         response_text = tokenizer.decode(response[0], skip_special_tokens=True)
         return response_text
 
-bot = ChatBot()
+# Create the Falcon chatbot instance
+StarlingBot_bot = StarlingBot()
+
+starling_bot = StarlingBot()  # Renamed for consistency
 
-title = "👋🏻Welcome to Tonic's Claire Chat🚀"
-description = "You can use this Space to test out the current model ([ClaireLLM](https://huggingface.co/OpenLLM-France/Claire-Mistral-7B-0.1)) or duplicate this Space and use it for any other model on 🤗HuggingFace. Join me on [Discord to build together](https://discord.gg/nXx5wbX9)."
-examples = [["[Estragon :] On va voir. Tiens. Ils prennent chacun un bout de la corde et tirent. La corde se casse. Ils manquent de tomber.", "[Vladimir] Fais voir quand même. (Estragon dénoue la corde qui maintient son pantalon.Celui-ci, beaucoup trop large, lui tombe autour des chevilles. Ils regardent la corde.) À la rigueur ça pourrait aller. Mais est-elle solide ?"]]
 iface = gr.Interface(
-    fn=bot.predict,
+    fn=starling_bot.predict,  # Corrected to match the instance name
     title=title,
     description=description,
     examples=examples,
     inputs=[
-        gr.Textbox(label="Deuxieme partie d'un dialogue"),
-        gr.Textbox(label="Premiere partie d'un dialogue")
+        gr.Textbox(label="User Message", type="text", lines=5),
+        gr.Textbox(label="💫🌠Starling Assistant Message or Instructions ", lines=2),
+        gr.Textbox(label="💫🌠Starling System Prompt or Instruction", lines=2),
+        gr.Checkbox(label="Advanced", value=False),  # Ensure this is connected to functionality
+        gr.Slider(label="Temperature", value=0.7, minimum=0.05, maximum=1.0, step=0.05),
+        gr.Slider(label="Max new tokens", value=100, minimum=25, maximum=256, step=1),
+        gr.Slider(label="Top-p (nucleus sampling)", value=0.90, minimum=0.01, maximum=0.99, step=0.05),
+        gr.Slider(label="Repetition penalty", value=1.9, minimum=1.0, maximum=2.0, step=0.05)
     ],
-    outputs=gr.outputs.Textbox(label="Claire LLM Dialogue"),
+    outputs="text",
     theme="ParityError/Anime"
 )
 
-iface.launch()