feat: address syntactically comments

integration-tests/conftest.py

@@ -374,14 +374,16 @@ def generate_load():
         prompt: str,
         max_new_tokens: int,
         n: int,
-        **kwargs,
+        seed: Optional[int] = None,
+        grammar: Optional[str] = None,
     ) -> List[Response]:
         futures = [
             client.generate(
                 prompt,
                 max_new_tokens=max_new_tokens,
                 decoder_input_details=True,
-                **kwargs,
+                seed=seed,
+                grammar=grammar,
             )
             for _ in range(n)
         ]

server/text_generation_server/utils/logits_process.py

@@ -1,6 +1,7 @@
 import math
 import torch
 
+from loguru import logger
 from functools import lru_cache
 from typing import Optional, List, Dict, Union
 
@@ -511,7 +512,7 @@ class GrammarLogitProcessor(LogitsProcessor):
         is_json_string = schema.startswith("{") and schema.endswith("}")
         regex_string = build_regex_from_object(schema) if is_json_string else schema
         fsm = RegexFSM(regex_string, tokenizer)
-        print(f"Compile FSM: {time.time() - start_time}")
+        logger.info(f"Compiled FSM in {time.time() - start_time:.2f}s")
         return fsm
 
     def adapt_tokenizer(self, tokenizer):
@@ -550,7 +551,7 @@ class HeterogeneousGrammarLogitProcessor(LogitsProcessor):
         self.tokenizer = tokenizer
 
     def __call__(self, input_ids: torch.Tensor, logits, fsm_grammar_states, grammars):
-        for i in range(len(logits)):
+        for i in range(logits.shape[0]):
             if fsm_grammar_states[i] == -1:
                 # todo mask for only eos token
                 continue
@@ -584,7 +585,7 @@ class HeterogeneousGrammarLogitProcessor(LogitsProcessor):
         is_json_string = schema.startswith("{") and schema.endswith("}")
         regex_string = build_regex_from_object(schema) if is_json_string else schema
         fsm = RegexFSM(regex_string, tokenizer)
-        # print(f"Compile FSM: {time.time() - start_time}")
+        logger.info(f"Compiled FSM in {time.time() - start_time:.2f}s")
         return fsm
 
     def adapt_tokenizer(self, tokenizer):

server/text_generation_server/utils/tokens.py

@@ -26,19 +26,19 @@ from transformers import PreTrainedTokenizerBase, RepetitionPenaltyLogitsProcess
 class NextTokenChooser:
     def __init__(
         self,
-        watermark=False,
-        temperature=1.0,
-        repetition_penalty=1.0,
-        frequency_penalty=0.0,
-        top_k=None,
-        top_p=None,
-        typical_p=None,
-        do_sample=False,
-        seed=0,
-        device="cpu",
-        tokenizer=None,
-        grammar=None,
-        fsm_grammar_state=None,
+        watermark: bool = False,
+        temperature: float = 1.0,
+        repetition_penalty: float = 1.0,
+        frequency_penalty: float = 0.0,
+        top_k: Optional[int] = None,
+        top_p: Optional[float] = None,
+        typical_p: Optional[float] = None,
+        do_sample: bool = False,
+        seed: int = 0,
+        device: str = "cpu",
+        tokenizer: Optional[PreTrainedTokenizerBase] = None,
+        grammar: str = "",
+        fsm_grammar_state: Optional[DefaultDict[int, int]] = None,
     ):
         self.watermark_processor = (
             WatermarkLogitsProcessor(device=device) if watermark else None
@@ -54,9 +54,7 @@ class NextTokenChooser:
             else None
         )
         self.grammar_processor = (
-            GrammarLogitProcessor(tokenizer, device)
-            if grammar and grammar != ""
-            else None
+            GrammarLogitProcessor(tokenizer, device) if grammar != "" else None
         )
         self.tokenizer = tokenizer
 
@@ -438,7 +436,10 @@ class HeterogeneousNextTokenChooser:
         self.grammars = new_grammars
         self.fsm_grammar_states = new_fsm_grammar_states
 
-        self.choice = Greedy()
+        if any(self.do_sample):
+            self.choice.filter(indices)
+        else:
+            self.choice = Greedy()
 
         return self
 
@@ -486,6 +487,7 @@ class Greedy:
     def __call__(self, logits):
         return logits.argmax(dim=-1)
 
+
 class HeterogeneousSampling:
     r"""
     Mixed greedy and probabilistic sampling. Compute both and pick the right one for each sample.