[style] Extract init-static values in tokenizer + multimodal path (#30709)

2026-07-09 19:36:34 -07:00
parent dda61b476e
commit ccd2028def
5 changed files with 49 additions and 40 deletions
@@ -262,6 +262,11 @@ class TokenizerManager(TokenizerControlMixin, TokenizerManagerScoreMixin):
        # Parse args
        self.server_args = server_args
        self.enable_metrics = server_args.enable_metrics
+        self.incremental_streaming_output = server_args.incremental_streaming_output
+        self.enable_lora = server_args.enable_lora
+        self.enable_trace = server_args.enable_trace
+        self.allow_auto_truncate = server_args.allow_auto_truncate
+        self.skip_tokenizer_init = server_args.skip_tokenizer_init
        self.preferred_sampling_params = server_args.preferred_sampling_params
        self.crash_dump_folder = server_args.crash_dump_folder
        set_global_server_args_for_tokenizer(server_args)
@@ -956,7 +961,7 @@ class TokenizerManager(TokenizerControlMixin, TokenizerManagerScoreMixin):

        # Validate input length
        if input_token_num >= self.context_len:
-            if self.server_args.allow_auto_truncate:
+            if self.allow_auto_truncate:
                logger.warning(
                    f"The input ({input_token_num} tokens) is longer than the "
                    f"model's context length ({self.context_len} tokens). "
@@ -977,7 +982,7 @@ class TokenizerManager(TokenizerControlMixin, TokenizerManagerScoreMixin):
            and max_new_tokens is not None
            and (max_new_tokens + input_token_num) > _max_req_len
        ):
-            if self.server_args.allow_auto_truncate:
+            if self.allow_auto_truncate:
                logger.warning(
                    f"Requested token count ({input_token_num} input + {max_new_tokens} new) "
                    f"exceeds the model's context length ({self.context_len} tokens). "
@@ -1432,7 +1437,7 @@ class TokenizerManager(TokenizerControlMixin, TokenizerManagerScoreMixin):
                del self.rid_to_state[state.obj.rid]

            # Mark ongoing LoRA request as finished.
-            if self.server_args.enable_lora and state.obj.lora_path:
+            if self.enable_lora and state.obj.lora_path:
                await self.lora_registry.release(state.obj.lora_id)
            if not is_stream:
                raise fastapi.HTTPException(
@@ -1479,9 +1484,7 @@ class TokenizerManager(TokenizerControlMixin, TokenizerManagerScoreMixin):

            # With incremental streaming, each chunk is a delta — coalesce
            # multiple queued chunks to avoid dropping token ids.
-            incremental_stream = (
-                is_stream and self.server_args.incremental_streaming_output
-            )
+            incremental_stream = is_stream and self.incremental_streaming_output
            if incremental_stream and len(out_list) > 1:
                out = self._coalesce_streaming_chunks(
                    out_list,
@@ -1906,8 +1909,7 @@ class TokenizerManager(TokenizerControlMixin, TokenizerManagerScoreMixin):
                    state,
                    state.obj.top_logprobs_num,
                    state.obj.token_ids_logprob,
-                    state.obj.return_text_in_logprobs
-                    and not self.server_args.skip_tokenizer_init,
+                    state.obj.return_text_in_logprobs and not self.skip_tokenizer_init,
                    recv_obj,
                    i,
                )
@@ -1972,9 +1974,7 @@ class TokenizerManager(TokenizerControlMixin, TokenizerManagerScoreMixin):
            if isinstance(recv_obj, BatchStrOutput):
                # Not all request types have `stream` (e.g., EmbeddingReqInput). Default to non-streaming.
                is_stream = getattr(state.obj, "stream", False)
-                incremental = (
-                    self.server_args.incremental_streaming_output and is_stream
-                )
+                incremental = is_stream and self.incremental_streaming_output
                delta_text = recv_obj.output_strs[i]
                delta_output_ids = list(recv_obj.output_ids[i])
                output_offset = state.last_output_offset
@@ -2022,9 +2022,7 @@ class TokenizerManager(TokenizerControlMixin, TokenizerManagerScoreMixin):
                    out_dict["prompt_token_ids"] = state.prompt_token_ids
            elif isinstance(recv_obj, BatchTokenIDOutput):
                is_stream = getattr(state.obj, "stream", False)
-                incremental = (
-                    self.server_args.incremental_streaming_output and is_stream
-                )
+                incremental = is_stream and self.incremental_streaming_output
                delta_output_ids = list(recv_obj.output_ids[i])
                output_offset = state.last_output_offset
                state.output_ids.extend(delta_output_ids)
@@ -2113,7 +2111,7 @@ class TokenizerManager(TokenizerControlMixin, TokenizerManagerScoreMixin):
                del self.rid_to_state[rid]

                # Mark ongoing LoRA request as finished.
-                if self.server_args.enable_lora and state.obj.lora_path:
+                if self.enable_lora and state.obj.lora_path:
                    asyncio.create_task(self.lora_registry.release(state.obj.lora_id))

            if out_dict is not None:
@@ -2779,7 +2777,7 @@ class TokenizerManager(TokenizerControlMixin, TokenizerManagerScoreMixin):
        if not obj.lora_path:
            return

-        if not self.server_args.enable_lora:
+        if not self.enable_lora:
            first_adapter = (
                obj.lora_path
                if isinstance(obj.lora_path, str)
@@ -2856,7 +2854,7 @@ class TokenizerManager(TokenizerControlMixin, TokenizerManagerScoreMixin):
        created_time = obj.received_time

        external_trace_header = None
-        if self.server_args.enable_trace:
+        if self.enable_trace:
            if obj.external_trace_header:
                # When the request comes from the rust grpc server or Engine there isn't a
                # real request object but we still need to propagate the trace context from
@@ -2889,7 +2887,7 @@ class TokenizerManager(TokenizerControlMixin, TokenizerManagerScoreMixin):
            time_stats = APIServerReqTimeStats(disagg_mode=self.disaggregation_mode)
            state = ReqState([], False, asyncio.Event(), sub_obj, time_stats)
            self.rid_to_state[rid] = state
-            if self.server_args.enable_trace:
+            if self.enable_trace:
                time_stats.init_trace_ctx(rid, bootstrap_room, external_trace_header)
            time_stats.set_created_time(created_time)

@@ -2962,7 +2960,7 @@ class TokenizerManager(TokenizerControlMixin, TokenizerManagerScoreMixin):
                    "mooncake",
                ]:
                    time_stats_json = None
-                    if self.server_args.enable_trace:
+                    if self.enable_trace:
                        state = self.rid_to_state.get(obj.rid)
                        if state is not None:
                            time_stats_json = state.time_stats.encode_json()
@@ -2986,7 +2984,7 @@ class TokenizerManager(TokenizerControlMixin, TokenizerManagerScoreMixin):
        """Convert attributes to span attributes."""
        span_attrs = {}

-        if not self.server_args.enable_trace:
+        if not self.enable_trace:
            return span_attrs

        # Token usage attributes
@@ -188,6 +188,9 @@ class BaseMultimodalProcessor(ABC):
        self._processor = _processor
        self.server_args = server_args
        self.transport_mode = transport_mode
+        self.keep_mm_feature_on_device = server_args.keep_mm_feature_on_device
+        self.disable_fast_image_processor = server_args.disable_fast_image_processor
+        self.skip_tokenizer_init = server_args.skip_tokenizer_init

        mm_process_config = self.server_args.mm_process_config
        self.image_config = mm_process_config.get("image", {})
@@ -436,7 +439,7 @@ class BaseMultimodalProcessor(ABC):
        if (
            hasattr(processor, "image_processor")
            and isinstance(processor.image_processor, BaseImageProcessor)
-            and not self.server_args.disable_fast_image_processor
+            and not self.disable_fast_image_processor
        ):
            if _is_cpu or get_server_args().rl_on_policy_target is not None:
                kwargs["device"] = "cpu"
@@ -470,7 +473,7 @@ class BaseMultimodalProcessor(ABC):
            return_tensors="pt",
            **kwargs,
        )
-        if not self.server_args.keep_mm_feature_on_device:
+        if not self.keep_mm_feature_on_device:
            # move feature tensors to cpu
            for feature_name in self.FEATURE_NAMES:
                if SGL_USE_CUDA_IPC:
@@ -832,7 +835,7 @@ class BaseMultimodalProcessor(ABC):

        # For MiniCPMO and MiniCPMV or multimodal_tokens not totally align, legacy show path
        if (
-            self.server_args.skip_tokenizer_init
+            self.skip_tokenizer_init
            or cnt[Modality.IMAGE] != n_image
            or cnt[Modality.VIDEO] != n_video
            or cnt[Modality.AUDIO] != n_audio
@@ -1234,7 +1237,7 @@ class BaseMultimodalProcessor(ABC):
                pool_byte_offset=byte_offset,
                pool_device_index=self.cudaipc_mmfeature_pool._pool_device_index,
            )
-        if self.server_args.keep_mm_feature_on_device:
+        if self.keep_mm_feature_on_device:
            return tensor
        return tensor.cpu()

@@ -303,7 +303,7 @@ class Ernie4_5_VLImageProcessor(SGLangBaseProcessor):
        if (
            hasattr(processor, "image_processor")
            and isinstance(processor.image_processor, BaseImageProcessor)
-            and not self.server_args.disable_fast_image_processor
+            and not self.disable_fast_image_processor
        ):
            if not _is_npu:
                kwargs["device"] = "cuda"
@@ -349,7 +349,7 @@ class Ernie4_5_VLImageProcessor(SGLangBaseProcessor):
                    if result["pixel_values_videos"].numel() == 0:
                        del result["pixel_values_videos"]

-        if not self.server_args.keep_mm_feature_on_device:
+        if not self.keep_mm_feature_on_device:
            # move feature tensors to cpu
            for feature_name in self.FEATURE_NAMES:
                if SGL_USE_CUDA_IPC:
@@ -289,6 +289,11 @@ class QwenVLImageProcessor(SGLangBaseProcessor):
        self.audio_start_token_id = getattr(hf_config, "audio_start_token_id", None)
        self.audio_token_id = getattr(hf_config, "audio_token_id", None)

+        self._spatial_merge_size = self.hf_config.vision_config.spatial_merge_size
+        self._tokens_per_second = getattr(
+            self.hf_config.vision_config, "tokens_per_second", None
+        )
+
        self.mm_tokens = MultimodalSpecialTokens(
            image_token="<|vision_start|><|image_pad|><|vision_end|>",
            image_token_id=hf_config.image_token_id,
@@ -300,6 +305,10 @@ class QwenVLImageProcessor(SGLangBaseProcessor):
            audio_token_id=self.audio_token_id,
        ).build(_processor)

+    @property
+    def spatial_merge_size(self):
+        return self._spatial_merge_size
+
    def build_input_ids_with_timestamps(
        self, prompt, embeddings, img_grid_thw, video_grid_thw, video_timestamps
    ):
@@ -311,7 +320,7 @@ class QwenVLImageProcessor(SGLangBaseProcessor):

        img_token_id = getattr(self, "IM_TOKEN_ID", None)
        video_token_id = getattr(self, "VIDEO_TOKEN_ID", None)
-        spatial_merge_size = getattr(self, "spatial_merge_size", 1)
+        spatial_merge_size = self.spatial_merge_size
        vision_start_token_id = getattr(self, "vision_start_token_id", None)
        vision_end_token_id = getattr(self, "vision_end_token_id", None)

@@ -401,14 +410,12 @@ class QwenVLImageProcessor(SGLangBaseProcessor):

        input_ids_tensor = torch.tensor(input_ids, dtype=torch.long).unsqueeze(0)
        mrope_positions, mrope_position_delta = MRotaryEmbedding.get_rope_index(
-            spatial_merge_size=self.hf_config.vision_config.spatial_merge_size,
+            spatial_merge_size=self._spatial_merge_size,
            image_token_id=self.mm_tokens.image_token_id,
            video_token_id=self.mm_tokens.video_token_id,
            vision_start_token_id=self.vision_start_token_id,
            model_type=self.model_type,
-            tokens_per_second=getattr(
-                self.hf_config.vision_config, "tokens_per_second", None
-            ),
+            tokens_per_second=self._tokens_per_second,
            input_ids=input_ids_tensor,
            image_grid_thw=image_grid_thw,
            video_grid_thw=video_grid_thw,
@@ -474,7 +481,7 @@ class QwenVLImageProcessor(SGLangBaseProcessor):
        if not image_items or len(image_items) != len(mm_items):
            return None

-        spatial_merge_size = self.hf_config.vision_config.spatial_merge_size
+        spatial_merge_size = self._spatial_merge_size
        sorted_items = sorted(image_items, key=lambda item: item.offsets[0][0])
        position_segments = []
        st = 0
@@ -615,7 +622,7 @@ class QwenVLImageProcessor(SGLangBaseProcessor):
        assert all(isinstance(modality, Modality) for modality in modality_list)

        mrope_positions, mrope_position_delta = MRotaryEmbedding.get_rope_index(
-            spatial_merge_size=self.hf_config.vision_config.spatial_merge_size,
+            spatial_merge_size=self._spatial_merge_size,
            image_token_id=self.mm_tokens.image_token_id,
            video_token_id=self.mm_tokens.video_token_id,
            vision_start_token_id=self.vision_start_token_id,
@@ -633,9 +640,7 @@ class QwenVLImageProcessor(SGLangBaseProcessor):
            position_id_per_seconds=getattr(
                self.hf_config, "position_id_per_seconds", None
            ),
-            tokens_per_second=getattr(
-                self.hf_config.vision_config, "tokens_per_second", None
-            ),
+            tokens_per_second=self._tokens_per_second,
        )
        mrope_positions = mrope_positions.squeeze(1)

@@ -783,14 +788,12 @@ class QwenVLImageProcessor(SGLangBaseProcessor):
                )
        if mrope_result is None:
            mrope_result = MRotaryEmbedding.get_rope_index(
-                spatial_merge_size=self.hf_config.vision_config.spatial_merge_size,
+                spatial_merge_size=self._spatial_merge_size,
                image_token_id=self.mm_tokens.image_token_id,
                video_token_id=self.mm_tokens.video_token_id,
                vision_start_token_id=self.vision_start_token_id,
                model_type=self.model_type,
-                tokens_per_second=getattr(
-                    self.hf_config.vision_config, "tokens_per_second", None
-                ),
+                tokens_per_second=self._tokens_per_second,
                # use the expanded token ids
                input_ids=input_ids.unsqueeze(0),
                image_grid_thw=image_grid_thw,