fix(server): fix index out of range for watermarking (#110)

server/text_generation_server/models/causal_lm.py

@@ -73,7 +73,7 @@ class CausalLMBatch(Batch):
             inputs.append(r.inputs)
             input_lengths.append(r.input_length)
             next_token_choosers.append(
-                NextTokenChooser.from_pb(r.parameters, len(tokenizer), device)
+                NextTokenChooser.from_pb(r.parameters, device)
             )
             stopping_criteria = StoppingCriteria.from_pb(
                 r.stopping_parameters, tokenizer

server/text_generation_server/models/galactica.py

@@ -103,7 +103,7 @@ class GalacticaCausalLMBatch(CausalLMBatch):
             inputs.append(escape_custom_split_sequence(r.inputs))
             input_lengths.append(r.input_length)
             next_token_choosers.append(
-                NextTokenChooser.from_pb(r.parameters, len(tokenizer), device)
+                NextTokenChooser.from_pb(r.parameters, device)
             )
             stopping_criteria = StoppingCriteria.from_pb(
                 r.stopping_parameters, tokenizer

server/text_generation_server/models/seq2seq_lm.py

@@ -83,7 +83,7 @@ class Seq2SeqLMBatch(Batch):
             decoder_input_ids.append(tokenizer.bos_token_id)
             decoder_input_lengths.append(1)
             next_token_choosers.append(
-                NextTokenChooser.from_pb(r.parameters, len(tokenizer), device)
+                NextTokenChooser.from_pb(r.parameters, device)
             )
             stopping_criteria = StoppingCriteria.from_pb(
                 r.stopping_parameters, tokenizer

server/text_generation_server/utils/tokens.py

@@ -36,7 +36,6 @@ class Greedy:
 class NextTokenChooser:
     def __init__(
         self,
-        vocab_size,
         watermark=False,
         temperature=1.0,
         repetition_penalty=1.0,
@@ -52,7 +51,7 @@ class NextTokenChooser:
         sampling = do_sample
 
         if watermark:
-            warpers.append(WatermarkLogitsProcessor(vocab_size, device=device))
+            warpers.append(WatermarkLogitsProcessor(device=device))
         if repetition_penalty is not None and repetition_penalty != 1.0:
             warpers.append(RepetitionPenaltyLogitsProcessor(penalty=repetition_penalty))
         if temperature is not None and temperature != 1.0:
@@ -85,11 +84,9 @@ class NextTokenChooser:
     def from_pb(
         cls,
         pb: generate_pb2.NextTokenChooserParameters,
-        vocab_size: int,
         device: torch.device,
     ) -> "NextTokenChooser":
         return NextTokenChooser(
-            vocab_size=vocab_size,
             watermark=pb.watermark,
             temperature=pb.temperature,
             repetition_penalty=pb.repetition_penalty,

server/text_generation_server/utils/watermark.py

@@ -25,14 +25,12 @@ DELTA = os.getenv("WATERMARK_DELTA", 2.0)
 class WatermarkLogitsProcessor(LogitsProcessor):
     def __init__(
         self,
-        vocab_size: int,
         gamma: float = GAMMA,
         delta: float = DELTA,
         hash_key: int = 15485863,  # just a large prime number to create a rng seed with sufficient bit width
         device: str = "cpu",
     ):
         # watermarking parameters
-        self.vocab_size = vocab_size
         self.gamma = gamma
         self.delta = delta
         self.rng = torch.Generator(device=device)
@@ -45,13 +43,13 @@ class WatermarkLogitsProcessor(LogitsProcessor):
         prev_token = input_ids[-1].item()
         self.rng.manual_seed(self.hash_key * prev_token)
 
-    def _get_greenlist_ids(self, input_ids: torch.LongTensor) -> list[int]:
+    def _get_greenlist_ids(self, input_ids: torch.LongTensor, max_value: int) -> list[int]:
         # seed the rng using the previous tokens/prefix
         self._seed_rng(input_ids)
 
-        greenlist_size = int(self.vocab_size * self.gamma)
+        greenlist_size = int(max_value * self.gamma)
         vocab_permutation = torch.randperm(
-            self.vocab_size, device=input_ids.device, generator=self.rng
+            max_value, device=input_ids.device, generator=self.rng
         )
         greenlist_ids = vocab_permutation[:greenlist_size]
         return greenlist_ids
@@ -76,7 +74,7 @@ class WatermarkLogitsProcessor(LogitsProcessor):
         self, input_ids: torch.LongTensor, scores: torch.FloatTensor
     ) -> torch.FloatTensor:
         assert len(input_ids) == 1
-        greenlist_ids = self._get_greenlist_ids(input_ids[0])
+        greenlist_ids = self._get_greenlist_ids(input_ids[0], scores.shape[-1])
         green_tokens_mask = self._calc_greenlist_mask(
             scores=scores, greenlist_token_ids=greenlist_ids
         )