fix llm

Author: ckrause

.gitignore

@@ -70,7 +70,7 @@ venv.bak/
 
 # Test models
 /sample_model/
-/test_model/
+/test_model*/
 
 # Training data
 /programs

loda/llm/data_preprocessing.py

@@ -149,7 +149,7 @@ def create_training_examples(self, max_examples: int = -1) -> List[TrainingExamp
 
             try:
                 # Read the program file
-                program_path = self.program_cache._get_path(program_id)
+                program_path = self.program_cache.path(program_id)
                 if not os.path.exists(program_path):
                     continue
 
@@ -171,12 +171,12 @@ def create_training_examples(self, max_examples: int = -1) -> List[TrainingExamp
 
                 # Validate that the code parses correctly
                 try:
-                    Program.parse(clean_code)
+                    Program(clean_code)
                 except Exception:
                     continue  # Skip programs that don't parse
 
                 example = TrainingExample(
-                    sequence_id=program_id,
+                    sequence_id=str(program_id),
                     description=description,
                     loda_code=clean_code,
                     terms=terms
@@ -229,7 +229,7 @@ def augment_descriptions(self, examples: List[TrainingExample]) -> List[Training
             # Create new examples for each variation
             for variation in variations:
                 augmented_example = TrainingExample(
-                    sequence_id=example.sequence_id + "_aug",
+                    sequence_id=str(example.sequence_id) + "_aug",
                     description=variation,
                     loda_code=example.loda_code,
                     terms=example.terms

loda/llm/inference.py

@@ -98,7 +98,7 @@ def _validate_and_evaluate_code(self, description: str, code: str) -> Generation
 
         try:
             # Try to parse the program
-            program = Program.parse(code)
+            program = Program(code)
 
             # Try to evaluate it for a few terms
             interpreter = Interpreter(max_memory=100, max_stack=10, max_steps=10000)

requirements.txt

@@ -9,3 +9,5 @@ transformers>=4.20.0
 datasets>=2.0.0
 tqdm>=4.62.0
 scikit-learn>=1.0.0
+sentencepiece>=0.1.95
+protobuf>=3.19.0