Some dead code cleanup

syncode/infer.py

@@ -23,18 +23,29 @@ class Syncode:
     Args:
         mode (str, optional): Mode for inference. Defaults to "grammar_mask". 
             "original" for original model, "grammar_mask" for grammar mask, "grammar_strict" for strict grammar mask.
+        
         model (str): Model id for huggingface model hub or model name if stored locally.
+        
         quantize (bool, optional): Quantize model. Defaults to True.
+        
         device (str, optional): Device to use. Defaults to "cuda".
+        
         num_samples (int, optional): Number of samples. Defaults to 1.
+        
         grammar (str, optional): Language. Defaults to "input". "input" is used for user input. 
             other options currently supported are "python", "go", "calc"
+        
         dataset (str, optional): Dataset. Defaults to "humaneval".
+        
         new_mask_store (bool, optional): Use new DFA mask store. Defaults to False.
+        
         num_few_shot (int, optional): Number of examples for few shot prompting. Defaults to 0.
+        
         chat_mode (bool, optional): Parse only the (output) and not (prompt+output) in chat mode. Defaults to False.
+        
         dev_mode (bool, optional): Development mode. Defaults to False.
         log_level (int, optional): Log level. Defaults to 2. 0 for no logs, 1 for minimal logs, 2 for all logs including time.
+        
         parser (str, optional): Parser to use. Defaults to "lalr".
         task_id (int, optional): For debugging a specific task. Defaults to None.
     """
@@ -109,8 +120,8 @@ def __init__(
                 mode=mode,
                 )
 
-        # Set LLM generation args e.g. max_new_tokens, do_sample, etc.
-        self.set_generation_args(kwargs, tokenizer)
+        # Set LLM max new tokens to 200 by default
+        kwargs['max_new_tokens'] = kwargs.get('max_new_tokens', 200)
 
         self.model = HuggingFaceModel(
             model, 
@@ -152,18 +163,6 @@ def get_output_path(self):
         out_path = out_dir + 'samples_' + str(self.num_samples) + '_mode_' + str(self.mode) + "_eval.jsonl"
         os.makedirs(out_dir, exist_ok=True)
         return out_dir,out_path
-
-    def set_generation_args(self, kwargs, tokenizer):
-        kwargs['max_new_tokens'] = kwargs.get('max_new_tokens', 200)
-        kwargs['do_sample'] = kwargs.get('do_sample', False)
-        kwargs['use_cache'] = kwargs.get('use_cache', True)
-        kwargs['eos_token_id'] = kwargs.get('eos_token_id', tokenizer.eos_token_id)
-        kwargs['pad_token_id'] = kwargs.get('pad_token_id', tokenizer.eos_token_id) # model has no pad token
-        if kwargs['do_sample'] or self.num_samples > 1: # If sampling, set temperature, top_k, top_p
-            kwargs['temperature'] = kwargs.get('temperature', 0.2)
-            kwargs['top_k'] = kwargs.get('top_k', self.num_samples)
-            kwargs['top_p'] = kwargs.get('top_p', 0.95)
-            print(f"Generation args: {kwargs}")
 
     def user_input(self, prompt:str, stop_words=[]):
         """

syncode/language_model.py

@@ -38,7 +38,6 @@ def __init__(
             logger: common.Logger, 
             tokenizer=None, 
             prompt_template: str = '', 
-            api_key: str = None, 
             best_of: int = 1, 
             before_prediction_hook=lambda: None, 
             device='cuda', 
@@ -67,7 +66,7 @@ def get_grammar_decoder(self):
         return None
 
     @torch.inference_mode()
-    def generate_batch_completion_grammar(self, prompt, batch_size, stop_words=[]) -> Iterable[str]:
+    def generate_batch_completion_grammar(self, prompt, batch_size, stop_words=None) -> Iterable[str]:
         '''
         Generates batch_size completions for the given prompt. 
         '''
@@ -87,7 +86,7 @@ def generate_batch_completion_grammar(self, prompt, batch_size, stop_words=[]) -
         gen_mode = self._get_generation_mode(gen_config)
 
         # Create stopping criteria
-        if len(stop_words) > 0:
+        if stop_words is not None:
             stop_criteria = StoppingCriteriaList([KeywordsStoppingCriteria(self.tokenizer, stop_words=stop_words)])
         else:
             stop_criteria = []
@@ -103,13 +102,11 @@ def generate_batch_completion_grammar(self, prompt, batch_size, stop_words=[]) -
                 )
         else:
             # Use generate from transformers library for other modes
-            if stop_criteria is not None:
-                print('Warning: Stopping criteria is not supported for batch generation')
-
             generated_ids = self.model.generate(
                 **inputs, 
                 logits_processor=self.logit_processors, 
-                stoppings=stop_words,
+                stop_strings=stop_words,
+                tokenizer=self.tokenizer,
                 **self.gen_args
                 )
         batch_completions = []