Rename compute_topk_logprobs to compute_topk_scores

The function now returns either logits or logprobs depending on
logits_mode, so "logprobs" in the name is misleading. "scores" is
the neutral term already used throughout the codebase.

Signed-off-by: Allen Shen <aoshen@inferact.ai>

Author: aoshen02

vllm/model_executor/models/diffusion_gemma.py

@@ -52,7 +52,7 @@
 from vllm.v1.worker.gpu.buffer_utils import UvaBackedTensor, async_copy_to_gpu
 from vllm.v1.worker.gpu.input_batch import InputBatch
 from vllm.v1.worker.gpu.model_states.interface import ModelState
-from vllm.v1.worker.gpu.sample.logprob import compute_topk_logprobs
+from vllm.v1.worker.gpu.sample.logprob import compute_topk_scores
 from vllm.v1.worker.gpu.sample.output import SamplerOutput
 from vllm.v1.worker.gpu.sample.penalties import use_penalty
 from vllm.v1.worker.gpu.states import RequestState
@@ -1343,7 +1343,7 @@ def __call__(
                     # positions are never emitted.
                     k_i = int(valid_canvas_len_np[li])
                     start = li * CL
-                    self._pending_logprobs[slot.item()] = compute_topk_logprobs(
+                    self._pending_logprobs[slot.item()] = compute_topk_scores(
                         flat_logits[start : start + k_i],
                         max_num_logprobs,
                         argmax_tokens[local_idx][:k_i],

vllm/v1/worker/gpu/sample/logprob.py

@@ -106,7 +106,7 @@ def compute_token_logprobs(
     return logprobs
 
 
-def compute_topk_logprobs(
+def compute_topk_scores(
     logits: torch.Tensor,
     num_logprobs: int,
     sampled_token_ids: torch.Tensor,

vllm/v1/worker/gpu/sample/prompt_logprob.py

@@ -9,7 +9,7 @@
 from vllm.triton_utils import tl, triton
 from vllm.v1.outputs import LogprobsTensors
 from vllm.v1.worker.gpu.input_batch import InputBatch
-from vllm.v1.worker.gpu.sample.logprob import compute_topk_logprobs
+from vllm.v1.worker.gpu.sample.logprob import compute_topk_scores
 
 
 class PromptLogprobsWorker:
@@ -229,7 +229,7 @@ def compute_prompt_logprobs_with_chunking(
         )
         chunk_tgt = prompt_token_ids[start_idx:end_idx]
 
-        result = compute_topk_logprobs(
+        result = compute_topk_scores(
             prompt_logits,
             requested_num,
             chunk_tgt,

vllm/v1/worker/gpu/sample/sampler.py

@@ -19,7 +19,7 @@
 from vllm.v1.worker.gpu.sample.logit_bias import LogitBiasState
 from vllm.v1.worker.gpu.sample.logprob import (
     LogprobTokenIdsState,
-    compute_topk_logprobs,
+    compute_topk_scores,
 )
 from vllm.v1.worker.gpu.sample.output import SamplerOutput
 from vllm.v1.worker.gpu.sample.penalties import PenaltiesState
@@ -105,7 +105,7 @@ def __call__(
             expanded_logits = logits.shape[0] != idx_mapping_np.shape[0]
             cu_num_logits = cu_num_logits_np.tolist() if expanded_logits else None
             num_logprobs = max_num_logprobs if max_num_logprobs != NO_LOGPROBS else 0
-            logprobs_tensors = compute_topk_logprobs(
+            logprobs_tensors = compute_topk_scores(
                 logits,
                 num_logprobs,
                 sampled,

vllm/v1/worker/gpu/spec_decode/rejection_sampler.py

@@ -11,7 +11,7 @@
     get_num_sampled_and_rejected,
 )
 from vllm.v1.worker.gpu.metrics.logits import get_num_nans
-from vllm.v1.worker.gpu.sample.logprob import compute_topk_logprobs
+from vllm.v1.worker.gpu.sample.logprob import compute_topk_scores
 from vllm.v1.worker.gpu.sample.output import SamplerOutput
 from vllm.v1.worker.gpu.sample.sampler import Sampler
 from vllm.v1.worker.gpu.sample.states import NO_LOGPROBS
@@ -91,7 +91,7 @@ def _get_logprobs_tensors(
             num_warps=1,
         )
         expanded_logits = num_logits != input_batch.idx_mapping.shape[0]
-        return compute_topk_logprobs(
+        return compute_topk_scores(
             logits,
             max_num_logprobs,
             flat_sampled,