no-repeat-ngram is processor not warper

server/text_generation_server/utils/logits_process.py

@@ -18,21 +18,15 @@ from transformers import (
     TopKLogitsWarper,
     TopPLogitsWarper,
     TypicalLogitsWarper,
-    NoRepeatNGramLogitsProcessor
 )
 
+from transformers.generation.logits_process import _calc_banned_ngram_tokens
+
 mempool = torch.cuda.graph_pool_handle() if torch.cuda.is_available() else None
 
 
 class StaticWarper:
-    def __init__(
-        self,
-        temperature=1.0,
-        top_k=None,
-        top_p=None,
-        typical_p=None,
-        no_repeat_ngram_size=None,
-    ):
+    def __init__(self, temperature=1.0, top_k=None, top_p=None, typical_p=None):
         self.warpers = []
 
         if temperature is not None and temperature != 1.0:
@@ -44,8 +38,6 @@ class StaticWarper:
             self.warpers.append(TopPLogitsWarper(top_p=top_p))
         if typical_p is not None and typical_p < 1.0:
             self.warpers.append(TypicalLogitsWarper(mass=typical_p))
-        if no_repeat_ngram_size is not None and no_repeat_ngram_size > 0:
-            self.warpers.append(NoRepeatNGramLogitsProcessor(no_repeat_ngram_size))
 
         self.cuda_graph = None
         self.static_scores = None
@@ -86,10 +78,12 @@ def static_warper(
     top_k: Optional[int],
     top_p: Optional[float],
     typical_p: Optional[float],
-    no_repeat_ngram_size: Optional[int],
 ) -> StaticWarper:
     return StaticWarper(
-        temperature=temperature, top_k=top_k, top_p=top_p, typical_p=typical_p, no_repeat_ngram_size=no_repeat_ngram_size
+        temperature=temperature,
+        top_k=top_k,
+        top_p=top_p,
+        typical_p=typical_p,
     )
 
 

server/text_generation_server/utils/tokens.py

@@ -19,7 +19,11 @@ from text_generation_server.utils.logits_process import (
     static_warper,
 )
 from text_generation_server.utils.watermark import WatermarkLogitsProcessor
-from transformers import PreTrainedTokenizerBase, RepetitionPenaltyLogitsProcessor
+from transformers import (
+    PreTrainedTokenizerBase,
+    RepetitionPenaltyLogitsProcessor,
+    NoRepeatNGramLogitsProcessor,
+)
 
 
 class NextTokenChooser:
@@ -59,6 +63,12 @@ class NextTokenChooser:
             if grammar != ""
             else None
         )
+
+        self.no_repeat_ngram_processor = (
+            NoRepeatNGramLogitsProcessor(no_repeat_ngram_size)
+            if no_repeat_ngram_size and no_repeat_ngram_size > 0
+            else None
+        )
         self.tokenizer = tokenizer
 
         has_warpers = (
@@ -66,11 +76,10 @@ class NextTokenChooser:
             or (top_k is not None and top_k != 0)
             or (top_p is not None and top_p < 1.0)
             or (typical_p is not None and typical_p < 1.0)
-            or (no_repeat_ngram_size is not None and no_repeat_ngram_size > 0)
         )
         if has_warpers:
             self.static_warper = static_warper(
-                temperature=temperature, top_k=top_k, top_p=top_p, typical_p=typical_p, no_repeat_ngram_size=no_repeat_ngram_size
+                temperature=temperature, top_k=top_k, top_p=top_p, typical_p=typical_p
             )
         else:
             self.static_warper = None
@@ -90,6 +99,8 @@ class NextTokenChooser:
             scores = self.frequency_processor(input_ids, scores)
         if self.grammar_processor is not None:
             scores = self.grammar_processor(scores, self.fsm_grammar_state)
+        if self.no_repeat_ngram_processor is not None:
+            scores = self.no_repeat_ngram_processor(input_ids, scores)
 
         if self.static_warper is None:
             next_logprob = torch.log_softmax(scores, -1)
@@ -243,6 +254,7 @@ class HeterogeneousNextTokenChooser:
         temperature: List[float],
         repetition_penalty: List[float],
         frequency_penalty: List[float],
+        no_repeat_ngram_size: List[int],
         top_k: List[int],
         top_p: List[float],
         typical_p: List[float],
@@ -291,6 +303,18 @@ class HeterogeneousNextTokenChooser:
             else None
         )
 
+        self.no_repeat_ngram_processor = (
+            HeterogeneousProcessorWrapper(
+                {
+                    i: NoRepeatNGramLogitsProcessor(n)
+                    for i, n in enumerate(no_repeat_ngram_size)
+                    if n > 0
+                }
+            )
+            if any([n > 0 for n in no_repeat_ngram_size])
+            else None
+        )
+
         if any(x != 1.0 for x in temperature):
             do_sample = [
                 sample or x != 1.0 for x, sample in zip(temperature, do_sample)
@@ -357,6 +381,8 @@ class HeterogeneousNextTokenChooser:
                 _scores = self.frequency_processor(input_ids, _scores)
             if self.grammar_processor is not None:
                 _scores = self.grammar_processor(_scores, self.fsm_grammar_states)
+            if self.no_repeat_ngram_processor is not None:
+                _scores = self.no_repeat_ngram_processor(input_ids, _scores)
             for warper in self.warpers:
                 _scores = warper(input_ids, _scores)
             _next_ids = self.choice(_scores)
@@ -491,6 +517,7 @@ class HeterogeneousNextTokenChooser:
             temperature=[pb_.temperature for pb_ in pb],
             repetition_penalty=[pb_.repetition_penalty for pb_ in pb],
             frequency_penalty=[pb_.frequency_penalty for pb_ in pb],
+            no_repeat_ngram_size=[pb_.no_repeat_ngram_size for pb_ in pb],
             top_k=[pb_.top_k for pb_ in pb],
             top_p=[pb_.top_p for pb_ in pb],
             typical_p=[pb_.typical_p for pb_ in pb],