From 63d9ad0a3a435cdf3a44495028b10f390a38f960 Mon Sep 17 00:00:00 2001 From: Matt Mastracci Date: Fri, 18 Sep 2026 14:11:27 -0600 Subject: [PATCH] [Model] DiffusionGemma: honor logprob_token_ids on the converging step (#57417) Signed-off-by: Matt Mastracci Co-authored-by: Claude Fable 5.1 --- vllm/model_executor/models/diffusion_gemma.py | 30 +++++++++++++++++-- 1 file changed, 27 insertions(+), 3 deletions(-) diff --git a/vllm/model_executor/models/diffusion_gemma.py b/vllm/model_executor/models/diffusion_gemma.py index 558bc82bcb..401d990870 100644 --- a/vllm/model_executor/models/diffusion_gemma.py +++ b/vllm/model_executor/models/diffusion_gemma.py @@ -967,6 +967,7 @@ class DiffusionSampler: tp_group_name: str = "", ): self.sampling_states = sampler.sampling_states + self.logprob_token_ids_state = sampler.logprob_token_ids_state self.req_states = sampler.req_states self.logits_mode = sampler.logprobs_mode in ("raw_logits", "processed_logits") # Self-conditioning soft embed = probs @ embed_weight * normalizer, @@ -1023,9 +1024,11 @@ class DiffusionSampler: # that was aborted between its converging denoise and commit steps. self._pending_logprobs.pop(req_idx, None) self.sampling_states.add_request(req_idx, sampling_params) + self.logprob_token_ids_state.add_request(req_idx, sampling_params) def apply_staged_writes(self) -> None: self.sampling_states.apply_staged_writes() + self.logprob_token_ids_state.apply_staged_writes() @property def penalties_state(self): @@ -1214,6 +1217,11 @@ class DiffusionSampler: slots_np = input_batch.idx_mapping_np[:num_reqs] max_num_logprobs = self.sampling_states.max_num_logprobs(slots_np) + # Requests may ask for specific token ids' logprobs instead of, or as + # well as, a top-k. + max_token_ids = self.logprob_token_ids_state.max_num_token_ids(slots_np) + want_logprobs = max_num_logprobs >= 0 or max_token_ids > 0 + num_logprobs = max(max_num_logprobs, 0) # Sample over the [num_decode * CL, vocab] logits. The fp32 pipeline in # _compiled_sample_step keeps several live [group * CL, vocab] copies, so @@ -1271,7 +1279,7 @@ class DiffusionSampler: # Logprobs for denoise steps that just converged (is_encoder_phase # flipped False→True), stashed per tile so `scaled` is freed each tile. - if max_num_logprobs >= 0: + if want_logprobs: converged_mask = states.is_encoder_phase[tile_slots] just_converged = converged_mask & ~is_committing[tile] if just_converged.any(): @@ -1285,17 +1293,33 @@ class DiffusionSampler: # positions are never emitted. k_i = int(valid_canvas_len_np[start_req + li]) pos = li * CL + per_req_ids = max_token_ids > 0 self._pending_logprobs[slot.item()] = compute_topk_scores( flat_logits[pos : pos + k_i], - max_num_logprobs, + num_logprobs, argmax_tokens[local_idx][:k_i], + logprob_token_ids_state=( + self.logprob_token_ids_state if per_req_ids else None + ), + # every row of this stash belongs to one slot + expanded_idx_mapping=( + torch.full( + (k_i,), + slot.item(), + dtype=torch.int32, + device=flat_logits.device, + ) + if per_req_ids + else None + ), + max_per_req_token_ids=max_token_ids, logits_mode=self.logits_mode, ) # Commit steps: is_committing was True at entry. Reassemble previously # stashed logprobs and attach to SamplerOutput. logprobs_tensors = None - if max_num_logprobs >= 0 and is_committing.any() and self._pending_logprobs: + if want_logprobs and is_committing.any() and self._pending_logprobs: committing_slots = set( decode_slots_np[is_committing.cpu().numpy()].tolist() )