Initial commit to add unique prefix - increasing number

Author: MML-coder

src/guidellm/dataset/synthetic.py

@@ -138,6 +138,8 @@ def __init__(
         self.text_creator = EndlessTextCreator(
             data=config.source,
         )
+        # Add counter for unique prefixes
+        self.request_counter = 0
 
     def __iter__(
         self,
@@ -170,22 +172,44 @@ def __iter__(
             output_tokens_sampler,
         ):
             start_index = rand.randint(0, len(self.text_creator.words))
+            # Increment counter for each request
+            self.request_counter += 1
             yield {
-                "prompt": self._create_prompt(prompt_tokens, start_index),
+                "prompt": self._create_prompt(prompt_tokens, start_index, self.request_counter),
                 "prompt_tokens_count": prompt_tokens,
                 "output_tokens_count": output_tokens,
             }
 
-    def _create_prompt(self, prompt_tokens: int, start_index: int) -> str:
+    def _create_prompt(self, prompt_tokens: int, start_index: int, request_id: int) -> str:
+        """
+        Create a prompt with unique prefix to prevent vLLM prefix caching.
+
+        Args:
+            prompt_tokens: Target number of tokens for the prompt
+            start_index: Starting position in the text corpus
+            request_id: Unique identifier for this request (used as prefix)
+        Returns:
+            Generated prompt string with unique prefix
+        """
         if prompt_tokens <= 0:
-            return ""
+            return f"{request_id}: "
+
+        # Create unique prefix that will prevent any prefix caching
+        unique_prefix = f"{request_id}: "
+
+        # Calculate how many tokens the prefix uses
+        prefix_tokens = len(self.processor.tokenize(unique_prefix))
+
+        # Adjust target tokens to account for the prefix
+        remaining_tokens = max(1, prompt_tokens - prefix_tokens)
 
         left = start_index
-        right = start_index + 4 * prompt_tokens
+        right = start_index + 4 * remaining_tokens
 
         while left < right:
             mid = (left + right) // 2
-            test_prompt = self.text_creator.create_text(start_index, mid - start_index)
+            base_text = self.text_creator.create_text(start_index, mid - start_index)
+            test_prompt = unique_prefix + base_text
             test_tokens = len(self.processor.tokenize(test_prompt))
 
             if test_tokens == prompt_tokens:
@@ -195,7 +219,8 @@ def _create_prompt(self, prompt_tokens: int, start_index: int) -> str:
             else:
                 right = mid
 
-        return self.text_creator.create_text(start_index, left - start_index)
+        base_text = self.text_creator.create_text(start_index, left - start_index)
+        return unique_prefix + base_text
 
 
 class SyntheticDatasetCreator(DatasetCreator):