Start work on llm_engine

sia/llm_engine.py

@@ -0,0 +1,72 @@
+from typing import NamedTuple
+from transformers import AutoTokenizer, AutoModelForCausalLM, pipeline, TextStreamer
+import torch
+
+class InferenceResult(NamedTuple):
+    reasoning: str
+    actions: str
+
+class LlmEngine:
+    def __init__(self, model_path: str):
+        """
+        Initialize the LLM Engine with a model path.
+
+        Args:
+            model_path: Path to the model weights to be used.
+        """
+        self.device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
+        self.torch_dtype=torch.float16 if torch.cuda.is_available() else torch.float32
+        print(f"device: {self.device}")
+        self.set_model_path(model_path)
+
+    def set_model_path(self, model_path: str):
+        """
+        Load the model from the specified path.
+
+        Args:
+            model_path: Path to the model weights to load.
+        """
+        tokenizer = AutoTokenizer.from_pretrained(model_path)
+        model = AutoModelForCausalLM.from_pretrained(
+            model_path,
+            return_dict=True,
+            low_cpu_mem_usage=True,
+            torch_dtype=self.torch_dtype,
+            device_map="auto",
+            trust_remote_code=True,
+        ).to(self.device)
+        if tokenizer.pad_token_id is None:
+            tokenizer.pad_token_id = tokenizer.eos_token_id
+        if model.config.pad_token_id is None:
+            model.config.pad_token_id = model.config.eos_token_id
+        self.pipeline = pipeline(
+            "text-generation",
+            model=model,
+            tokenizer=tokenizer,
+            torch_dtype=torch.float16,
+            device_map="auto",
+        )
+
+    def infer(self, system_prompt: str, main_context: str, action_schema: str) -> InferenceResult:
+        """
+        Run inference using the system prompt and main context, while validating actions against the provided XML schema.
+
+        Args:
+            system_prompt: The system prompt string
+            main_context: The main context string after templating
+            action_schema: XML schema to validate the generated actions
+
+        Returns:
+            InferenceResult: the actions validate against the schema
+        """
+        pass
+
+    def finetune(self, dataset_paths: list, output_dir: str):
+        """
+        Fine-tune the model with new datasets and save the updated model weights.
+
+        Args:
+            dataset_paths: List of paths to datasets for fine-tuning.
+            output_dir: Directory where the updated model weights will be saved.
+        """
+        pass