[V1][Core] min_p sampling support

Signed-off-by: Aoyu <[email protected]>

Author: AoyuQC

tests/v1/sample/test_sampler.py

@@ -81,6 +81,8 @@ def _create_default_sampling_metadata(
         top_k=torch.empty(batch_size, ),
         no_top_p=True,
         no_top_k=True,
+        min_p=torch.empty(batch_size, ),
+        no_min_p=True,
         generators={},
         max_num_logprobs=0,
         prompt_token_ids=_create_prompt_tokens_tensor(prompt_token_ids,
@@ -338,6 +340,44 @@ def test_sampler_repetition_penalty(device: str, batch_size: int,
 
 @pytest.mark.parametrize("device", CUDA_DEVICES)
 @pytest.mark.parametrize("batch_size", [1, 2, 32])
+@pytest.mark.parametrize("min_p", [0.0, 0.1])
+def test_sampler_min_p(device: str, batch_size: int, min_p: float):
+    """
+    Tests that when min_p is applied, tokens with probability below 
+    min_p * max_prob are masked with -inf.
+    """
+    torch.set_default_device(device)
+    fake_logits = _create_fake_logits(batch_size, VOCAB_SIZE)
+
+    # Create one dominant token per batch
+    for i in range(batch_size):
+        fake_logits[i, 0] = 10.0  # High logit for first token
+        fake_logits[i, 1:] = 1e-2  # Others remain low
+
+    sampling_metadata = _create_default_sampling_metadata(
+        NUM_OUTPUT_TOKENS, batch_size, VOCAB_SIZE, torch.device(device))
+
+    # Configure min_p parameters
+    sampling_metadata.min_p = torch.full((batch_size, ), min_p, device=device)
+
+    sampler = Sampler()
+    logits = sampler.apply_min_p(fake_logits, sampling_metadata.min_p)
+    logits = logits.cpu()
+
+    for batch_idx in range(batch_size):
+        for token_id in range(VOCAB_SIZE):
+            if token_id == 0:
+                # Dominant token should always be unmasked
+                assert logits[batch_idx][token_id] != -float("inf")
+            else:
+                if min_p > 0.0:
+                    # Non-dominant tokens should be masked when min_p > 0
+                    assert logits[batch_idx][token_id] == -float("inf")
+                else:
+                    # No masking when min_p is 0
+                    assert logits[batch_idx][token_id] != -float("inf")
+
+
 @pytest.mark.parametrize("bias_value", [-0.1, 1.2])
 def test_sampler_logit_bias(device: str, batch_size: int, bias_value: float):
     """

vllm/v1/sample/metadata.py

@@ -17,6 +17,8 @@ class SamplingMetadata:
     top_k: torch.Tensor
     no_top_p: bool
     no_top_k: bool
+    min_p: torch.Tensor
+    no_min_p: bool
 
     generators: Dict[int, torch.Generator]
 

vllm/v1/sample/sampler.py

@@ -93,6 +93,10 @@ def sample(
             sampling_metadata.no_top_p,
             sampling_metadata.top_p,
         )
+
+        if not sampling_metadata.no_min_p:
+            logits = self.apply_min_p(logits, sampling_metadata.min_p)
+
         if sampling_metadata.all_random:
             return random_sampled
 
@@ -169,6 +173,27 @@ def apply_penalties(
                 sampling_metadata.output_token_ids)
         return logits
 
+    def apply_min_p(
+        self,
+        logits: torch.Tensor,
+        min_p: torch.Tensor,
+    ) -> torch.Tensor:
+        """
+        Filters logits using adaptive probability thresholding.
+        """
+        # Convert logits to probability distribution
+        probability_values = torch.nn.functional.softmax(logits, dim=-1)
+        # Calculate maximum probabilities per sequence
+        max_probabilities = torch.amax(probability_values,
+                                       dim=-1,
+                                       keepdim=True)
+        # Reshape min_p for broadcasting
+        adjusted_min_p = min_p.unsqueeze(1) * max_probabilities
+        # Identify valid tokens using threshold comparison
+        valid_token_mask = probability_values >= adjusted_min_p
+        # Apply mask using boolean indexing
+        logits[~valid_token_mask] = -float('inf')
+
     def apply_logits_bias(
         self,
         logits: torch.Tensor,

vllm/v1/worker/gpu_input_batch.py

@@ -14,6 +14,8 @@
 from vllm.v1.sample.metadata import SamplingMetadata
 from vllm.v1.worker.block_table import BlockTable
 
+_SAMPLING_EPS = 1e-5
+
 if TYPE_CHECKING:
     from vllm.multimodal.inputs import PlaceholderRange
 
@@ -120,6 +122,16 @@ def __init__(
         self.top_k_cpu = self.top_k_cpu_tensor.numpy()
         self.top_k_reqs: Set[str] = set()
 
+        self.min_p = torch.empty((max_num_reqs, ),
+                                 dtype=torch.float32,
+                                 device=device)
+        self.min_p_cpu_tensor = torch.empty((max_num_reqs, ),
+                                            dtype=torch.float32,
+                                            device="cpu",
+                                            pin_memory=pin_memory)
+        self.min_p_cpu = self.min_p_cpu_tensor.numpy()
+        self.min_p_reqs: Set[str] = set()
+
         # Frequency penalty related data structures
         self.frequency_penalties = torch.empty((max_num_reqs, ),
                                                dtype=torch.float,
@@ -223,8 +235,11 @@ def add_request(
         self.top_k_cpu[req_index] = sampling_params.top_k
         if sampling_params.top_k > 0:
             self.top_k_reqs.add(req_id)
+        self.min_p_cpu[req_index] = sampling_params.min_p
         self.frequency_penalties_cpu[
             req_index] = sampling_params.frequency_penalty
+        if sampling_params.min_p > _SAMPLING_EPS:
+            self.min_p_reqs.add(req_id)
         if sampling_params.frequency_penalty != 0.0:
             self.frequency_penalties_reqs.add(req_id)
         self.presence_penalties_cpu[
@@ -273,6 +288,7 @@ def remove_request(self, req_id: str) -> Optional[int]:
         self.random_reqs.discard(req_id)
         self.top_p_reqs.discard(req_id)
         self.top_k_reqs.discard(req_id)
+        self.min_p_reqs.discard(req_id)
         self.frequency_penalties_reqs.discard(req_id)
         self.presence_penalties_reqs.discard(req_id)
         self.repetition_penalties_reqs.discard(req_id)
@@ -299,6 +315,7 @@ def clear(self) -> None:
         self.random_reqs.clear()
         self.top_p_reqs.clear()
         self.top_k_reqs.clear()
+        self.min_p_reqs.clear()
         self.frequency_penalties_reqs.clear()
         self.presence_penalties_reqs.clear()
         self.repetition_penalties_reqs.clear()
@@ -354,6 +371,7 @@ def condense(self, empty_req_indices: List[int]) -> None:
                 empty_index] = self.presence_penalties_cpu[last_req_index]
             self.repetition_penalties_cpu[
                 empty_index] = self.repetition_penalties_cpu[last_req_index]
+            self.min_p_cpu[empty_index] = self.min_p_cpu[last_req_index]
             self.min_tokens[empty_index] = self.min_tokens[last_req_index]
             self.stop_token_ids[empty_index] = self.stop_token_ids[
                 last_req_index]
@@ -381,6 +399,8 @@ def make_sampling_metadata(
                 self.top_p_cpu_tensor[:self.num_reqs], non_blocking=True)
             self.top_k[:self.num_reqs].copy_(
                 self.top_k_cpu_tensor[:self.num_reqs], non_blocking=True)
+            self.min_p[:self.num_reqs].copy_(
+                self.min_p_cpu_tensor[:self.num_reqs], non_blocking=True)
             if not self.no_penalties:
                 # Since syncing these tensors is expensive only copy them
                 # if necessary i.e. if there are requests which require
@@ -421,6 +441,8 @@ def make_sampling_metadata(
             all_random=self.all_random,
             top_p=self.top_p[:self.num_reqs],
             top_k=self.top_k[:self.num_reqs],
+            min_p=self.min_p[:self.num_reqs],
+            no_min_p=self.no_min_p,
             no_top_p=self.no_top_p,
             no_top_k=self.no_top_k,
             generators=self.generators,
@@ -497,6 +519,10 @@ def no_top_p(self) -> bool:
     def no_top_k(self) -> bool:
         return len(self.top_k_reqs) == 0
 
+    @property
+    def no_min_p(self) -> bool:
+        return len(self.min_p_reqs) == 0
+
     @property
     def no_penalties(self) -> bool:
         return (len(self.presence_penalties_reqs) == 0