[V1] Do not detokenize if sampling param detokenize is False (#14224)

Signed-off-by: Himanshu Jaju <hj@mistral.ai>
Signed-off-by: Nick Hill <nhill@redhat.com>
Co-authored-by: Nick Hill <nhill@redhat.com>

tests/v1/sample/test_sampling_params_e2e.py

@@ -14,7 +14,10 @@ PROMPT = "Hello my name is Robert and I"
 
 @pytest.fixture(scope="module")
 def model() -> LLM:
-    return LLM(MODEL, enforce_eager=True)
+    # Disable prefix caching so that we can test prompt logprobs.
+    # TODO remove this after https://github.com/vllm-project/vllm/pull/13949
+    # is merged
+    return LLM(MODEL, enforce_eager=True, enable_prefix_caching=False)
 
 
 def test_n_gt_1(model):
@@ -87,9 +90,33 @@ def test_stop_token_ids(model):
 
     stop_token_ids = [stop_token_id_0, stop_token_id_1]
     params = SamplingParams(temperature=0, stop_token_ids=stop_token_ids)
+    output = model.generate(PROMPT, params)
     assert output[0].outputs[0].token_ids[-1] == stop_token_id_0
 
 
+def test_detokenize_false(model):
+    """Check that detokenize=False option works."""
+
+    output = model.generate(PROMPT, SamplingParams(detokenize=False))
+    assert len(output[0].outputs[0].token_ids) > 0
+    assert len(output[0].outputs[0].text) == 0
+
+    output = model.generate(
+        PROMPT, SamplingParams(detokenize=False, logprobs=3,
+                               prompt_logprobs=3))
+    assert len(output[0].outputs[0].token_ids) > 0
+    assert len(output[0].outputs[0].text) == 0
+
+    prompt_logprobs = output[0].prompt_logprobs
+    sampled_logprobs = output[0].outputs[0].logprobs
+    assert len(prompt_logprobs) > 1
+    assert len(sampled_logprobs) > 1
+    for all_logprobs in (prompt_logprobs[1:], sampled_logprobs):
+        for logprobs in all_logprobs:
+            assert 3 <= len(logprobs) <= 4
+            assert all(lp.decoded_token is None for lp in logprobs.values())
+
+
 def test_bad_words(model):
     """Check that we respect bad words."""
 

vllm/v1/engine/detokenizer.py

@@ -1,6 +1,6 @@
 # SPDX-License-Identifier: Apache-2.0
 
-from dataclasses import dataclass
+from dataclasses import dataclass, field
 from typing import Optional
 
 from vllm.engine.output_processor.stop_checker import StopChecker
@@ -16,41 +16,46 @@ logger = init_logger(__name__)
 class IncrementalDetokenizer:
 
     # Generation data
-    output_text: str
-    tokens: list[str]
     token_ids: list[int]
-    prompt_len: int
+    output_text: str = ""
+    tokens: list[str] = field(default_factory=list)
+    prompt_len: int = 0
 
     # Stop strings
-    stop: list[str]
-    include_stop_str_in_output: bool
+    stop: list[str] = field(default_factory=list)
+    include_stop_str_in_output: bool = False
 
     # Metadata for incremental detokenization
-    prefix_offset: int
-    read_offset: int
+    prefix_offset: int = 0
+    read_offset: int = 0
 
     # Parameters for detokenization
-    skip_special_tokens: bool
-    spaces_between_special_tokens: bool
+    skip_special_tokens: bool = True
+    spaces_between_special_tokens: bool = True
 
-    # Tokenizer for this request
-    tokenizer: AnyTokenizer
+    # Tokenizer for this request,
+    # None if detokenization is disabled.
+    tokenizer: Optional[AnyTokenizer] = None
 
     # Accounting for stop string buffering
-    stop_buffer_length: int
+    stop_buffer_length: int = 0
     _last_output_text_offset: int = 0
 
     @property
     def output_token_ids(self) -> list[int]:
-        return self.token_ids[self.prompt_len:]
+        return self.token_ids if not self.prompt_len else (
+            self.token_ids[self.prompt_len:])
 
     @classmethod
     def from_new_request(
         cls,
-        tokenizer: AnyTokenizer,
+        tokenizer: Optional[AnyTokenizer],
         request: EngineCoreRequest,
     ) -> "IncrementalDetokenizer":
 
+        if tokenizer is None:
+            return cls(token_ids=[])
+
         tokens, prefix_offset, read_offset = convert_prompt_ids_to_tokens(
             tokenizer=tokenizer,
             prompt_ids=request.prompt_token_ids,
@@ -66,7 +71,6 @@ class IncrementalDetokenizer:
             stop_buffer_length = 0
 
         return cls(
-            output_text="",
             tokens=tokens,
             # Detokenizer mutates this list, so need a unique copy.
             # NOTE(Nick): could we take ownership of it though?
@@ -93,6 +97,10 @@ class IncrementalDetokenizer:
         Return matched stop string or None.
         """
 
+        if self.tokenizer is None:
+            self.token_ids.extend(new_token_ids)
+            return None
+
         # 1) Detokenize the new token ids incrementally.
         # TODO(woosuk): This method becomes very inefficient when the number of
         # new_token_ids is more than 1. We need to optimize this.

vllm/v1/engine/logprobs.py

@@ -1,6 +1,7 @@
 # SPDX-License-Identifier: Apache-2.0
 
 import itertools
+from collections.abc import Iterable
 from dataclasses import dataclass
 from typing import Optional
 
@@ -13,12 +14,15 @@ from vllm.v1.outputs import LogprobsLists, LogprobsTensors
 
 logger = init_logger(__name__)
 
+NONES = itertools.repeat(None)
+
 
 @dataclass
 class LogprobsProcessor:
 
-    # Tokenizer for this request
-    tokenizer: AnyTokenizer
+    # Tokenizer for this request,
+    # None if detokenization is disabled.
+    tokenizer: Optional[AnyTokenizer]
 
     # Logprobs for this request
     logprobs: Optional[SampleLogprobs]
@@ -30,7 +34,7 @@ class LogprobsProcessor:
     @classmethod
     def from_new_request(
         cls,
-        tokenizer: AnyTokenizer,
+        tokenizer: Optional[AnyTokenizer],
         request: EngineCoreRequest,
     ) -> "LogprobsProcessor":
         num_logprobs = request.sampling_params.logprobs
@@ -66,8 +70,8 @@ class LogprobsProcessor:
                                              token_ids_lst):
 
             # Detokenize (non-incrementally).
-            decoded_tokens = convert_ids_list_to_tokens(
-                self.tokenizer, token_ids)
+            decoded_tokens = NONES if self.tokenizer is None else (
+                convert_ids_list_to_tokens(self.tokenizer, token_ids))
 
             # Sampler puts the sampled logprob in first.
             sampled_token_logprob = logprobs[0]
@@ -103,9 +107,9 @@ class LogprobsProcessor:
 
         # Detokenize non-incrementally.
         # Output is flat: [num_tok, num_lps] -> [num_tok * num_lps]
-        decoded_tokens = convert_ids_list_to_tokens(
-            self.tokenizer,
-            token_ids.flatten().tolist())
+        decoded_tokens = None if self.tokenizer is None else (
+            convert_ids_list_to_tokens(self.tokenizer,
+                                       token_ids.flatten().tolist()))
 
         # Recover shapes.
         num_prompt_tokens, num_logprobs = logprobs.shape
@@ -121,7 +125,8 @@ class LogprobsProcessor:
             # Handle flattening.
             offset = pos * num_logprobs
             offset_end = offset + num_logprobs
-            decoded_tokens_for_pos = decoded_tokens[offset:offset_end]
+            decoded_tokens_for_pos = NONES \
+            if decoded_tokens is None else decoded_tokens[offset:offset_end]
 
             # Update with the Logprob dictionary for this pos.
             self.prompt_logprobs.append(
@@ -153,7 +158,7 @@ class LogprobsProcessor:
     def _make_logprob_dict(
         logprobs: list[float],
         logprob_token_ids: list[int],
-        decoded_tokens: list[str],
+        decoded_tokens: Iterable[Optional[str]],
         rank: int,
         num_logprobs: int,
     ) -> dict[int, Logprob]:

vllm/v1/engine/output_processor.py

@@ -68,6 +68,8 @@ class RequestState:
         queue: Optional[asyncio.Queue[RequestOutput]],
         log_stats: bool,
     ) -> "RequestState":
+        if not request.sampling_params.detokenize:
+            tokenizer = None
         return cls(
             request_id=request.request_id,
             parent_req=parent_req,