load new phi

handler.py

@@ -1,123 +1,33 @@
-from llama_cpp import Llama
-from typing import Dict, List, Any, Union
-import os
-
+from typing import Dict, List, Any
+from transformers import AutoModelForCausalLM, AutoTokenizer
+import torch
 
 class EndpointHandler:
-    _instance = None  # Singleton instance
-    _model_loaded = False  # Flag to check if the model is loaded
-
-    def __new__(cls, *args, **kwargs):  # Allow arguments to pass through
-        if not cls._instance:
-            cls._instance = super(EndpointHandler, cls).__new__(cls)
-        return cls._instance
-
-    def __init__(self, model_path=""):
-        if not self._model_loaded:
-            # Construct the model path assuming the model is in the same directory as the handler file
-            script_dir = os.path.dirname(os.path.abspath(__file__))
-            model_filename = "Phi-3-medium-128k-instruct-IQ2_XS.gguf"
-            self.model_path = os.path.join(script_dir, model_filename)
-
-            # Check if the model file exists
-            if not os.path.exists(self.model_path):
-                raise ValueError(f"Model path does not exist: {self.model_path}")
-
-            # Load the GGUF model using llama_cpp
-            self.llm = Llama(
-                model_path=self.model_path,
-                n_ctx=5000,  # Set context length to 5000 tokens
-                # n_threads=12,  # Adjust the number of CPU threads as per your machine
-                n_gpu_layers=-1  # Adjust based on GPU availability
-            )
+    def __init__(self, path=""):
+        # load model and processor from path
+        self.tokenizer = AutoTokenizer.from_pretrained("microsoft/Phi-3-medium-4k-instruct", trust_remote_code=True)
+        self.model = AutoModelForCausalLM.from_pretrained("microsoft/Phi-3-medium-4k-instruct", trust_remote_code=True)
 
-            # Define generation kwargs for the model
-            self.generation_kwargs = {
-                "max_tokens": 400,  # Respond with up to 400 tokens
-                "stop": ["<|end|>", "<|user|>", "<|assistant|>"],
-                "top_k": 1  # Greedy decoding
-            }
-
-            self._model_loaded = True
-            
-    @classmethod
-    def get_instance(cls, model_path=""):
-        """Provides access to the singleton instance."""
-        if not cls._instance:
-            cls._instance = cls(model_path)  # Create instance if it doesn't exist
-        return cls._instance
-
-    def __call__(self, data: Union[Dict[str, Any], str]) -> List[Dict[str, Any]]:
+    def __call__(self, data: Dict[str, Any]) -> Dict[str, str]:
         """
-        Data args:
-            inputs (:obj:`dict`): The input prompts for the LLM including system instructions and user messages.
-            str: A string input which will create a chat completion.
-
-        Return:
-            A :obj:`list` | `dict`: will be serialized and returned.
+        Args:
+            data (:obj:):
+                includes the deserialized image file as PIL.Image
         """
-        if isinstance(data, dict):
-            # Extract inputs
-            inputs = data.get("inputs", {})
-            system_instructions = inputs.get("system", "")
-            user_message = inputs.get("message", "")
-
-            if not user_message:
-                raise ValueError("No user message provided for the model.")
+        # process input
+        inputs = data.pop("inputs", data)
+        parameters = data.pop("parameters", None)
 
-            # Combine system instructions and user message
-            final_input = f"{system_instructions}\n{user_message}"
-
-            # Run inference with llama_cpp
-            response = self.llm.create_chat_completion(
-                messages=[
-                    {"role": "system", "content": system_instructions},
-                    {"role": "user", "content": user_message}
-                ],
-                **self.generation_kwargs
-            )
-
-        elif isinstance(data, str):
-            # Create a chat completion from the input string
-            response = self.llm.create_chat_completion(
-                messages=[
-                    {"role": "user", "content": data}
-                ],
-                **self.generation_kwargs
-            )
+        # preprocess
+        input_ids = self.tokenizer(inputs, return_tensors="pt").input_ids
 
+        # pass inputs with all kwargs in data
+        if parameters is not None:
+            outputs = self.model.generate(input_ids, **parameters)
         else:
-            raise ValueError("Invalid input type. Expected dict or str, got {}".format(type(data)))
-
-        # Access generated text based on the response structure
-        try:
-            generated_text = response["choices"][0]["message"].get("content", "")
-        except (KeyError, IndexError):
-            raise ValueError("Unexpected response structure: missing 'content' in 'choices[0]['message']'")
-
-        # Return the generated text
-        return [{"generated_text": generated_text}]
-
-
-def main():
-    handler = EndpointHandler()  # assume Handler is the class that contains the __call__ method
-
-    # Test 1: Dictionary input
-    data_dict = {"inputs": {"system": "System instructions", "message": "Hello, how are you?"}}
-    result_dict = handler(data_dict)
-    print("Dictionary input result:", result_dict)
-
-    # Test 2: String input
-    data_str = "Hello, how are you?"
-    result_str = handler(data_str)
-    print("String input result:", result_str)
+            outputs = self.model.generate(input_ids)
 
-    # Test 3: Invalid input type
-    data_invalid = 123
-    try:
-        handler(data_invalid)
-    except ValueError as e:
-        print("Invalid input type error:", e)
+        # postprocess the prediction
+        prediction = self.tokenizer.decode(outputs[0], skip_special_tokens=True)
 
-if __name__ == "__main__":
-    main()
+        return [{"generated_text": prediction}]

requirements.txt

@@ -1,3 +1 @@
-llama-cpp-python -C cmake.args="-DLLAMA_BLAS=ON;-DLLAMA_BLAS_VENDOR=OpenBLAS"
-torch
-transformers
+transformers>=4.4.0