Create app.py

app.py

@@ -0,0 +1,130 @@
+from pydantic import BaseModel
+from llama_cpp import Llama
+from concurrent.futures import ThreadPoolExecutor, as_completed
+import re
+import httpx
+import asyncio
+import gradio as gr
+import os
+from dotenv import load_dotenv
+import json
+
+load_dotenv()
+
+HUGGINGFACE_TOKEN = os.getenv("HUGGINGFACE_TOKEN")
+
+global_data = {
+    'models': {},
+    'tokens': {
+        'eos': 'eos_token',
+        'pad': 'pad_token',
+        'padding': 'padding_token',
+        'unk': 'unk_token',
+        'bos': 'bos_token',
+        'sep': 'sep_token',
+        'cls': 'cls_token',
+        'mask': 'mask_token'
+    }
+}
+
+model_configs = [
+    {"repo_id": "Hjgugugjhuhjggg/mergekit-ties-tzamfyy-Q2_K-GGUF", "filename": "mergekit-ties-tzamfyy-q2_k.gguf", "name": "my_model"}
+    # Add more models here as needed.  Ensure the repo_id and filename are correct.
+]
+
+class ModelManager:
+    def __init__(self):
+        self.models = {}
+
+    def load_model(self, model_config):
+        if model_config['name'] not in self.models:
+            try:
+                model = Llama.from_pretrained(repo_id=model_config['repo_id'], filename=model_config['filename'], use_auth_token=HUGGINGFACE_TOKEN)
+                self.models[model_config['name']] = model
+                print(f"Model '{model_config['name']}' loaded successfully.")
+            except Exception as e:
+                print(f"Error loading model {model_config['name']}: {e}")
+                self.models[model_config['name']] = None # Indicate loading failure
+
+
+    def load_all_models(self):
+        with ThreadPoolExecutor() as executor:
+            futures = [executor.submit(self.load_model, config) for config in model_configs]
+            for future in as_completed(futures):
+                future.result() # Propagate exceptions during loading
+        return self.models
+
+model_manager = ModelManager()
+global_data['models'] = model_manager.load_all_models()
+
+
+class ChatRequest(BaseModel):
+    message: str
+
+def normalize_input(input_text):
+    return input_text.strip()
+
+def remove_duplicates(text):
+    lines = text.split('\n')
+    unique_lines = []
+    seen_lines = set()
+    for line in lines:
+        line = line.strip() #Added to remove extra whitespace
+        if line and line not in seen_lines:
+            unique_lines.append(line)
+            seen_lines.add(line)
+    return '\n'.join(unique_lines)
+
+def generate_model_response(model, inputs):
+    try:
+        if model is None: #Handle failed model loading
+            return ""
+        response = model(inputs)
+        return remove_duplicates(response['choices'][0]['text'])
+    except Exception as e:
+        print(f"Error generating model response: {e}")
+        return f"Error: {e}" #Return informative error message
+
+def remove_repetitive_responses(responses):
+    unique_responses = {}
+    for response in responses:
+        if response['model'] not in unique_responses and response['response']: #added check for empty responses
+            unique_responses[response['model']] = response['response']
+    return unique_responses
+
+
+async def process_message(message, history):
+    inputs = normalize_input(message)
+    with ThreadPoolExecutor() as executor:
+        futures = [
+            executor.submit(generate_model_response, model, inputs)
+            for model in global_data['models'].values()
+        ]
+        responses = [{'model': model_name, 'response': future.result()} for model_name, future in zip(global_data['models'].keys(), as_completed(futures))]
+
+    unique_responses = remove_repetitive_responses(responses)
+    formatted_response = ""
+    for model, response in unique_responses.items():
+        formatted_response += f"**{model}:**\n{response}\n\n"
+
+    history.append((message, formatted_response))
+    return history, ""
+
+
+iface = gr.Interface(
+    fn=process_message,
+    inputs=[
+        gr.Textbox(lines=2, placeholder="Enter your message here..."),
+        gr.State([])
+    ],
+    outputs=[
+        gr.Chatbot(),
+        gr.Textbox(label="cURL command", visible=False) #Hidden cURL command
+    ],
+    title="Multi-Model LLM API",
+    description="Enter a message and get responses from multiple LLMs.",
+)
+
+if __name__ == "__main__":
+    port = int(os.environ.get("PORT", 7860))
+    iface.launch(server_port=port)