[V1][Bugfix][Spec Decode] Fix incorrect outputs in V1 speculative decoding due to batch indexing (#14645)

Signed-off-by: Benjamin Chislett <benjamin.chislett@centml.ai>

tests/v1/e2e/test_ngram_spec_decode.py

@@ -1,4 +1,6 @@
 # SPDX-License-Identifier: Apache-2.0
+import random
+
 import pytest
 
 from vllm import LLM, SamplingParams
@@ -6,16 +8,41 @@ from vllm import LLM, SamplingParams
 
 @pytest.fixture
 def test_prompts():
-    return [
+    prompt_types = ["repeat", "sentence"]
-        "Can you repeat the sentence ten times, this is a sentence.",
+    num_prompts = 100
-        "Can you repeat the sentence ten times, this is a test.",
+    prompts = []
-    ]
+
+    random.seed(0)
+    random_prompt_type_choices = random.choices(prompt_types, k=num_prompts)
+
+    # Generate a mixed batch of prompts, some of which can be easily
+    # predicted by n-gram matching and some which likely cannot.
+    for kind in random_prompt_type_choices:
+        word_choices = ["test", "temp", "hello", "where"]
+        word = random.choice(word_choices)
+        if kind == "repeat":
+            prompt = f"""
+            please repeat the word '{word}' 10 times.
+            give no other output than the word at least ten times in a row,
+            in lowercase with spaces between each word and without quotes.
+            """
+        elif kind == "sentence":
+            prompt = f"""
+            please give a ten-word sentence that
+            uses the word {word} at least once.
+            give no other output than that simple sentence without quotes.
+            """
+        else:
+            raise ValueError(f"Unknown prompt type: {kind}")
+        prompts.append([{"role": "user", "content": prompt}])
+
+    return prompts
 
 
 @pytest.fixture
 def sampling_config():
     # Only support greedy for now
-    return SamplingParams(temperature=0, max_tokens=30, ignore_eos=False)
+    return SamplingParams(temperature=0, max_tokens=10, ignore_eos=False)
 
 
 @pytest.fixture
@@ -32,18 +59,28 @@ def test_ngram_correctness(monkeypatch, test_prompts, sampling_config,
     with monkeypatch.context() as m:
         m.setenv("VLLM_USE_V1", "1")
 
-        ref_llm = LLM(model=model_name)
+        ref_llm = LLM(model=model_name, max_model_len=1024)
-        ref_outputs = ref_llm.generate(test_prompts, sampling_config)
+        ref_outputs = ref_llm.chat(test_prompts, sampling_config)
         del ref_llm
 
         spec_llm = LLM(model=model_name,
                        speculative_model='[ngram]',
                        ngram_prompt_lookup_max=5,
                        ngram_prompt_lookup_min=3,
-                       num_speculative_tokens=3)
+                       num_speculative_tokens=3,
-        spec_outputs = spec_llm.generate(test_prompts, sampling_config)
+                       max_model_len=1024)
+        spec_outputs = spec_llm.chat(test_prompts, sampling_config)
+        matches = 0
+        misses = 0
         for ref_output, spec_output in zip(ref_outputs, spec_outputs):
-            assert ref_output.outputs[0].text == spec_output.outputs[0].text, \
+            if ref_output.outputs[0].text == spec_output.outputs[0].text:
-                (f"ref_output: {ref_output.outputs[0].text},"
+                matches += 1
-                 f"spec_output: {spec_output.outputs[0].text}")
+            else:
+                misses += 1
+                print(f"ref_output: {ref_output.outputs[0].text}")
+                print(f"spec_output: {spec_output.outputs[0].text}")
+
+        # Heuristic: expect at least 70% of the prompts to match exactly
+        # Upon failure, inspect the outputs to check for inaccuracy.
+        assert matches > int(0.7 * len(ref_outputs))
         del spec_llm

vllm/v1/worker/gpu_model_runner.py

@@ -1015,11 +1015,9 @@ class GPUModelRunner(LoRAModelRunnerMixin):
         else:
             target_probs = self.model.sampler.compute_probs(
                 logits, sampling_metadata)
-            scheduled_request_ids = scheduler_output.num_scheduled_tokens.keys(
-            )
             draft_token_ids = [
                 scheduler_output.scheduled_spec_decode_tokens.get(req_id, [])
-                for req_id in scheduled_request_ids
+                for req_id in self.input_batch.req_ids
             ]
             sampler_output = self.rejection_sampler(draft_token_ids,
                                                     target_probs,