[Bugfix] Fix speculative decoding with MLPSpeculator with padded vocabulary (vllm-project#7218)

Signed-off-by: Travis Johnson <[email protected]>

Author: tjohnson31415

tests/spec_decode/e2e/test_mlp_correctness.py

@@ -19,8 +19,12 @@
 correctess for the target model outputs.
 """
 
+from unittest.mock import patch
+
 import pytest
 
+from vllm.model_executor.layers.vocab_parallel_embedding import pad_vocab_size
+
 from .conftest import (run_equality_correctness_test,
                        run_greedy_equality_correctness_test)
 
@@ -178,6 +182,62 @@ def test_mlp_e2e_greedy_correctness_with_preemption(baseline_llm_generator,
                                          force_output_len=True)
 
 
+@pytest.mark.parametrize(
+    "common_llm_kwargs",
+    [{
+        "block_size": 8,
+        # 2 for small prompt, 256//8 for generated.
+        "num_gpu_blocks_override": 2 + 256 // 8,
+        "max_model_len": (2 + 256 // 8) * 8,
+
+        # Skip cuda graph recording for fast test.
+        "enforce_eager": True,
+
+        # Required for spec decode.
+        "use_v2_block_manager": True,
+
+        # Precision
+        "dtype": PRECISION,
+
+        # Main model
+        "model": MAIN_MODEL,
+    }])
+@pytest.mark.parametrize("per_test_common_llm_kwargs", [{}])
+@pytest.mark.parametrize("baseline_llm_kwargs", [{}])
+@pytest.mark.parametrize("test_llm_kwargs", [
+    {
+        "speculative_model": SPEC_MODEL,
+    },
+])
+@pytest.mark.parametrize(
+    "output_len",
+    [
+        # Use small output len for fast test.
+        128,
+    ])
+@pytest.mark.parametrize("batch_size", [4])
+@pytest.mark.parametrize("seed", [1])
+def test_mlp_e2e_greedy_correctness_with_padding(baseline_llm_generator,
+                                                 test_llm_generator,
+                                                 batch_size: int,
+                                                 output_len: int):
+    """Verify greedy equality when the vocab dimension is padded
+    """
+
+    # Default pad_to is 64, test model has vocab_size of 32000
+    def patched_pad_vocab_size(vocab_size, pad_to=None):
+        return pad_vocab_size(vocab_size, pad_to=32064)
+
+    with patch(
+            "vllm.model_executor.layers.vocab_parallel_embedding.pad_vocab_size",
+            patched_pad_vocab_size):
+        run_greedy_equality_correctness_test(baseline_llm_generator,
+                                             test_llm_generator,
+                                             batch_size,
+                                             max_output_len=output_len,
+                                             force_output_len=True)
+
+
 @pytest.mark.parametrize(
     "common_llm_kwargs",
     [{

vllm/model_executor/layers/logits_processor.py

@@ -91,7 +91,7 @@ def _get_logits(self, hidden_states: torch.Tensor,
             logits = tensor_model_parallel_all_gather(logits)
         # Remove paddings in vocab (if any).
         if logits is not None:
-            logits = logits[:, :self.org_vocab_size]
+            logits = logits[..., :self.org_vocab_size]
         return logits
 
     def extra_repr(self) -> str:

vllm/model_executor/layers/rejection_sampler.py

@@ -78,8 +78,8 @@ def forward(
         # Only perform shape/dtype/device checking in strict mode, as it adds
         # overhead.
         if self._strict_mode:
-            self._raise_if_incorrect_input(target_probs, bonus_token_ids,
-                                           draft_probs, draft_token_ids)
+            self._raise_if_incorrect_input(target_probs, draft_token_ids,
+                                           bonus_token_ids, draft_probs)
 
         accepted, recovered_token_ids = (
             self._batch_modified_rejection_sampling(

vllm/model_executor/models/mlp_speculator.py

@@ -175,13 +175,14 @@ def generate_proposals(
             states.add_(z, alpha=self.emb_weight / self.state_weight)
 
             states = self.activation(self.ln[head_index](states))  # b k d
-            # TODO: not yet supporting top_k_tokens_per_head
             previous_hidden_states = states
+            # TODO: not yet supporting top_k_tokens_per_head
+            states = states.flatten(0, 1)
 
             logits = self.logits_processor(self.head[head_index], states,
                                            sampling_metadata)
 
-            output = self.sampler(logits.flatten(0, 1), sampling_metadata)
+            output = self.sampler(logits, sampling_metadata)
             last_tokens = output.sampled_token_ids
             next_tokens.append(output)