Update space

app.py

@@ -1,24 +1,141 @@
+# # import gradio as gr
+# # from huggingface_hub import InferenceClient
+
+# # """
+# # For more information on `huggingface_hub` Inference API support, please check the docs: https://huggingface.co/docs/huggingface_hub/v0.22.2/en/guides/inference
+# # """
+# # client = InferenceClient("HuggingFaceH4/zephyr-7b-beta")
+
+
+# # def respond(
+# #     message,
+# #     history: list[tuple[str, str]],
+# #     system_message,
+# #     max_tokens,
+# #     temperature,
+# #     top_p,
+# # ):
+# #     messages = [{"role": "system", "content": system_message}]
+
+# #     for val in history:
+# #         if val[0]:
+# #             messages.append({"role": "user", "content": val[0]})
+# #         if val[1]:
+# #             messages.append({"role": "assistant", "content": val[1]})
+
+# #     messages.append({"role": "user", "content": message})
+
+# #     response = ""
+
+# #     for message in client.chat_completion(
+# #         messages,
+# #         max_tokens=max_tokens,
+# #         stream=True,
+# #         temperature=temperature,
+# #         top_p=top_p,
+# #     ):
+# #         token = message.choices[0].delta.content
+
+# #         response += token
+# #         yield response
+
+
+# # """
+# # For information on how to customize the ChatInterface, peruse the gradio docs: https://www.gradio.app/docs/chatinterface
+# # """
+# # demo = gr.ChatInterface(
+# #     respond,
+# #     additional_inputs=[
+# #         gr.Textbox(value="You are a friendly Chatbot.", label="System message"),
+# #         gr.Slider(minimum=1, maximum=2048, value=512, step=1, label="Max new tokens"),
+# #         gr.Slider(minimum=0.1, maximum=4.0, value=0.7, step=0.1, label="Temperature"),
+# #         gr.Slider(
+# #             minimum=0.1,
+# #             maximum=1.0,
+# #             value=0.95,
+# #             step=0.05,
+# #             label="Top-p (nucleus sampling)",
+# #         ),
+# #     ],
+# # )
+
+
+# # if __name__ == "__main__":
+# #     demo.launch()
+
 import torch
 import gradio as gr
 from transformers import AutoModelForCausalLM, AutoTokenizer
+import os
+from safetensors.torch import load_file, save_file
+
+# Define model names
+# MODEL_1_PATH = "./adapter_model.safetensors"  # Local path inside Space
+###
+MODEL_1_PATH = "Priyanka6/fine-tuning-inference"
+###
+MODEL_2_NAME = "sarvamai/sarvam-1"  # The base model on Hugging Face Hub
+# MODEL_3_NAME = 
+
+def trim_adapter_weights(model_path):
+    """
+    Trims the last token from the adapter's lm_head.lora_B.default.weight 
+    if there is a mismatch with the base model.
+    """
+    model_path = "./adapter_model.safetensors"
+    # if not os.path.exists(model_path):
+    #     raise FileNotFoundError(f"Adapter file not found: {model_path}")
+    
+    checkpoint = load_file(model_path)
+    print("Keys in checkpoint:", list(checkpoint.keys()))
 
-# Load the model and tokenizer only once at startup
-MODEL_NAME = "sarvamai/sarvam-1"
-tokenizer = None
-model = None
-
-def load_model():
-    global tokenizer, model
-    if tokenizer is None or model is None:
-        tokenizer = AutoTokenizer.from_pretrained(MODEL_NAME)
-        model = AutoModelForCausalLM.from_pretrained(MODEL_NAME, torch_dtype=torch.float16, device_map="auto")
-        model.eval()
-
-def respond(message, history, max_tokens, temperature, top_p):
-    global tokenizer, model
-    # Ensure model is loaded
-    load_model()
+    key_to_trim = "lm_head.lora_B.default.weight"
     
+    if key_to_trim in checkpoint:
+        print("Entered")
+        original_size = checkpoint[key_to_trim].shape[0]
+        expected_size = original_size - 1  # Removing last token
+        
+        print(f"Trimming {key_to_trim}: {original_size} -> {expected_size}")
+
+        checkpoint[key_to_trim] = checkpoint[key_to_trim][:-1]  # Trim the last row
+
+        # Save the modified adapter
+        trimmed_adapter_path = os.path.join(model_path, "adapter_model_trimmed.safetensors")
+        save_file(checkpoint, trimmed_adapter_path)
+        return trimmed_adapter_path
+    print("did execute the if block")
+    return model_path
+model_path=os.path.join(MODEL_1_PATH,"adapter_model.safetensors")
+trimmed_adapter_path = trim_adapter_weights(model_path)
+
+# Load the tokenizer (same for both models)
+TOKENIZER_NAME = "sarvamai/sarvam-1"
+tokenizer = AutoTokenizer.from_pretrained(TOKENIZER_NAME)
+
+# Function to load a model
+def load_model(model_choice):
+    if model_choice == "Hugging face dataset":
+        model = AutoModelForCausalLM.from_pretrained("./", torch_dtype=torch.float16, device_map="auto")
+        trimmed_adapter_path = os.path.join("Priyanka6/fine-tuning-inference", "adapter_model_trimmed.safetensors")
+        model.load_adapter(trimmed_adapter_path, "safe_tensors")  # Load safetensors adapter
+    else:
+        model = AutoModelForCausalLM.from_pretrained(MODEL_2_NAME)
+    model.eval()
+    return model
+
+# Load default model on startup
+current_model = load_model("Hugging face dataset")
+
+# Chatbot response function
+def respond(message, history, model_choice, max_tokens, temperature, top_p):
+    global current_model
+    
+    # Switch model if user selects a different one
+    if (model_choice == "Hugging face dataset" and current_model is not None and current_model.config.name_or_path != MODEL_1_PATH) or \
+       (model_choice == "Proprietary dataset1" and current_model is not None and current_model.config.name_or_path != MODEL_2_NAME):
+        current_model = load_model(model_choice)
+
     # Convert chat history to format
     messages = [{"role": "system", "content": "You are a friendly AI assistant."}]
     for val in history:
@@ -31,11 +148,8 @@ def respond(message, history, max_tokens, temperature, top_p):
     # Tokenize and generate response
     inputs = tokenizer.apply_chat_template(messages, tokenize=False)
     input_tokens = tokenizer(inputs, return_tensors="pt").to("cuda" if torch.cuda.is_available() else "cpu")
-    
-    # Get the length of input tokens to separate new response
-    input_length = input_tokens.input_ids.shape[1]
 
-    output_tokens = model.generate(
+    output_tokens = current_model.generate(
         **input_tokens,
         max_new_tokens=max_tokens,
         temperature=temperature,
@@ -43,29 +157,30 @@ def respond(message, history, max_tokens, temperature, top_p):
         pad_token_id=tokenizer.pad_token_id,
         eos_token_id=tokenizer.eos_token_id,
     )
-    
-    # Extract only the new tokens (the model's response)
-    new_tokens = output_tokens[0][input_length:]
-    response = tokenizer.decode(new_tokens, skip_special_tokens=True)
-    
-    # Clean up any remaining system prompt or formatting artifacts
-    response = response.strip()
-    if response.startswith("assistant:"):
-        response = response[len("assistant:"):].strip()
-    
+
+    response = tokenizer.decode(output_tokens[0], skip_special_tokens=True)
     return response
 
 # Define Gradio Chat Interface
 demo = gr.ChatInterface(
     fn=respond,
     additional_inputs=[
+        gr.Dropdown(choices=["Hugging face dataset", "Proprietary dataset1"], value="Fine-Tuned Model", label="Select Model"),
         gr.Slider(minimum=1, maximum=1024, value=256, step=1, label="Max Tokens"),
         gr.Slider(minimum=0.1, maximum=2.0, value=0.7, step=0.1, label="Temperature"),
         gr.Slider(minimum=0.1, maximum=1.0, value=0.95, step=0.05, label="Top-p"),
     ],
-    title="Sarvam-1 Chat Interface",
-    description="Chat with the Sarvam-1 language model"
 )
 
 if __name__ == "__main__":
-    demo.launch()
+    demo.launch()
+
+
+# # Test the chatbot
+# if __name__ == "__main__":
+#     while True:
+#         query = input("User: ")
+#         if query.lower() in ["exit", "quit"]:
+#             break
+#         response = chat(query)
+#         print(f"Bot: {response}")