Revert "[Renderer] Separate out RendererConfig from ModelConfig (#30145)" (#30199)

This commit is contained in:
Cyrus Leung
2025-12-07 00:00:22 -08:00
committed by GitHub
parent 27f4c2fd46
commit e83b7e379c
105 changed files with 797 additions and 969 deletions
-2
View File
@@ -16,7 +16,6 @@ from vllm.config import (
LoadConfig,
ModelConfig,
ParallelConfig,
RendererConfig,
SchedulerConfig,
VllmConfig,
)
@@ -217,7 +216,6 @@ def create_vllm_config(
return VllmConfig(
model_config=model_config,
renderer_config=RendererConfig(model_config=model_config),
cache_config=cache_config,
parallel_config=parallel_config,
scheduler_config=scheduler_config,
+4 -16
View File
@@ -8,7 +8,7 @@ import pytest
import torch
import vllm.v1.core.kv_cache_utils as kv_cache_utils
from vllm.config import ModelConfig, RendererConfig, SchedulerConfig, VllmConfig
from vllm.config import ModelConfig, SchedulerConfig, VllmConfig
from vllm.lora.request import LoRARequest
from vllm.multimodal.inputs import (
MultiModalFeatureSpec,
@@ -667,10 +667,7 @@ def test_metrics_empty_stats():
def test_get_kv_cache_configs_multiple_workers():
model_config = ModelConfig(max_model_len=16)
vllm_config = VllmConfig(
model_config=model_config,
renderer_config=RendererConfig(model_config=model_config),
)
vllm_config = VllmConfig(model_config=model_config)
ref_kv_cache_spec = new_kv_cache_spec()
same_kv_cache_specs = [
@@ -1139,7 +1136,6 @@ def test_estimate_max_model_len(model_id, max_model_len, want_estimated_max_len)
vllm_config = VllmConfig(
model_config=model_config,
renderer_config=RendererConfig(model_config=model_config),
scheduler_config=scheduler_config,
)
@@ -1179,7 +1175,6 @@ def test_get_max_concurrency_for_kv_cache_config():
vllm_config = VllmConfig(
model_config=model_config,
renderer_config=RendererConfig(model_config=model_config),
scheduler_config=scheduler_config,
)
@@ -1298,10 +1293,7 @@ def test_allocate_with_lookahead():
def test_get_kv_cache_config_one_worker():
# pass max_model_len to pass check_enough_kv_cache_memory
model_config = ModelConfig(max_model_len=16)
vllm_config = VllmConfig(
model_config=model_config,
renderer_config=RendererConfig(model_config=model_config),
)
vllm_config = VllmConfig(model_config=model_config)
mem_per_block_per_layer = 16 * 2 * 64 * 4 * 2
# all layers are full attention -> single group
@@ -1592,11 +1584,7 @@ def test_get_kv_cache_config_one_worker():
def test_get_kv_cache_configs_attention_free():
kv_cache_specs: dict[str, KVCacheSpec] = {}
model_config = ModelConfig(max_model_len=16)
vllm_config = VllmConfig(
model_config=model_config,
renderer_config=RendererConfig(model_config=model_config),
)
vllm_config = VllmConfig(model_config=ModelConfig(max_model_len=16))
kv_cache_configs = get_kv_cache_configs(vllm_config, [kv_cache_specs], [0])
assert kv_cache_configs == [
KVCacheConfig(
-2
View File
@@ -11,7 +11,6 @@ from vllm.config import (
ECTransferConfig,
KVTransferConfig,
ModelConfig,
RendererConfig,
SchedulerConfig,
SpeculativeConfig,
VllmConfig,
@@ -1564,7 +1563,6 @@ def create_scheduler_with_priority(
vllm_config = VllmConfig(
scheduler_config=scheduler_config,
model_config=model_config,
renderer_config=RendererConfig(model_config=model_config),
cache_config=cache_config,
kv_transfer_config=kv_transfer_config,
speculative_config=speculative_config,
-2
View File
@@ -9,7 +9,6 @@ from vllm.config import (
ECTransferConfig,
KVTransferConfig,
ModelConfig,
RendererConfig,
SchedulerConfig,
SpeculativeConfig,
VllmConfig,
@@ -133,7 +132,6 @@ def create_scheduler(
vllm_config = VllmConfig(
scheduler_config=scheduler_config,
model_config=model_config,
renderer_config=RendererConfig(model_config=model_config),
cache_config=cache_config,
kv_transfer_config=kv_transfer_config,
speculative_config=speculative_config,
-2
View File
@@ -15,7 +15,6 @@ from vllm.config import (
ECTransferConfig,
KVTransferConfig,
ModelConfig,
RendererConfig,
SchedulerConfig,
VllmConfig,
)
@@ -523,7 +522,6 @@ def test_encoder_instance_zero_kv_cache(
vllm_config = VllmConfig(
model_config=model_config,
renderer_config=RendererConfig(model_config=model_config),
cache_config=cache_config,
scheduler_config=scheduler_config,
kv_transfer_config=kv_transfer_config,
@@ -5,14 +5,7 @@ import pytest
from vllm.assets.image import ImageAsset
from vllm.assets.video import VideoAsset
from vllm.config import (
CacheConfig,
DeviceConfig,
ModelConfig,
MultiModalConfig,
RendererConfig,
VllmConfig,
)
from vllm.config import CacheConfig, DeviceConfig, ModelConfig, VllmConfig
from vllm.sampling_params import SamplingParams
from vllm.v1.engine import input_processor as input_processor_mod
from vllm.v1.engine.input_processor import InputProcessor
@@ -51,21 +44,22 @@ def _mock_input_processor(
monkeypatch.setattr(VllmConfig, "__post_init__", lambda self: None, raising=True)
model_config = ModelConfig(
skip_tokenizer_init=True,
max_model_len=128,
mm_processor_cache_gb=mm_cache_gb,
generation_config="vllm",
)
model_config.multimodal_config = MultiModalConfig(mm_processor_cache_gb=mm_cache_gb)
renderer_config = RendererConfig(
model_config=model_config,
tokenizer="dummy",
skip_tokenizer_init=True,
)
# Minimal multimodal_config to satisfy references in
# Processor.process_inputs.
class _MockMMConfig:
def __init__(self, gb: float):
self.mm_processor_cache_gb = gb
model_config.multimodal_config = _MockMMConfig(mm_cache_gb) # type: ignore[attr-defined]
vllm_config = VllmConfig(
model_config=model_config,
renderer_config=renderer_config,
cache_config=CacheConfig(enable_prefix_caching=enable_prefix_caching),
device_config=DeviceConfig(device="cpu"),
)
-2
View File
@@ -15,7 +15,6 @@ from vllm.config import (
DeviceConfig,
KVTransferConfig,
ModelConfig,
RendererConfig,
SchedulerConfig,
VllmConfig,
)
@@ -128,7 +127,6 @@ def create_vllm_config(
return VllmConfig(
scheduler_config=scheduler_config,
model_config=model_config,
renderer_config=RendererConfig(model_config=model_config),
cache_config=cache_config,
kv_transfer_config=kv_transfer_config,
device_config=DeviceConfig("cpu"),
-2
View File
@@ -19,7 +19,6 @@ from vllm.config import (
DeviceConfig,
ModelConfig,
ParallelConfig,
RendererConfig,
SchedulerConfig,
SpeculativeConfig,
VllmConfig,
@@ -62,7 +61,6 @@ def _create_proposer(
vllm_config = VllmConfig(
model_config=model_config,
renderer_config=RendererConfig(model_config=model_config),
cache_config=CacheConfig(),
speculative_config=speculative_config,
device_config=DeviceConfig(device=current_platform.device_type),
-2
View File
@@ -18,7 +18,6 @@ from vllm.config import (
DeviceConfig,
ModelConfig,
ParallelConfig,
RendererConfig,
SchedulerConfig,
SpeculativeConfig,
VllmConfig,
@@ -47,7 +46,6 @@ def _create_mtp_proposer(num_speculative_tokens: int) -> EagleProposer:
vllm_config = VllmConfig(
model_config=model_config,
renderer_config=RendererConfig(model_config=model_config),
cache_config=CacheConfig(),
speculative_config=speculative_config,
device_config=DeviceConfig(device=current_platform.device_type),
-2
View File
@@ -4,7 +4,6 @@ import numpy as np
from vllm.config import (
ModelConfig,
RendererConfig,
SpeculativeConfig,
VllmConfig,
)
@@ -70,7 +69,6 @@ def test_ngram_proposer():
return NgramProposer(
vllm_config=VllmConfig(
model_config=model_config,
renderer_config=RendererConfig(model_config=model_config),
speculative_config=SpeculativeConfig(
prompt_lookup_min=min_n,
prompt_lookup_max=max_n,
@@ -6,7 +6,7 @@ from concurrent.futures import Future
import pytest
from transformers import AutoTokenizer
from vllm.config import RendererConfig, StructuredOutputsConfig, VllmConfig
from vllm.config import StructuredOutputsConfig, VllmConfig
from vllm.config.model import ModelConfig
from vllm.config.parallel import ParallelConfig
from vllm.config.speculative import SpeculativeConfig
@@ -72,11 +72,8 @@ def test_backend_guidance_rollback_terminated():
def test_grammar_bitmask_with_specdec():
tokenizer = AutoTokenizer.from_pretrained(TOKENIZER)
prompt = tokenizer.encode('{"a": "b"}')
model_config = ModelConfig(tokenizer=TOKENIZER)
vllm_config = VllmConfig(
model_config=model_config,
renderer_config=RendererConfig(model_config=model_config, tokenizer=TOKENIZER),
model_config=ModelConfig(tokenizer=TOKENIZER),
structured_outputs_config=StructuredOutputsConfig(backend="guidance"),
speculative_config=SpeculativeConfig(model="[ngram]", num_speculative_tokens=3),
)
@@ -140,11 +137,8 @@ def test_grammar_init_async_and_sync(async_grammar):
# Use "external_launcher" for sync mode, None for async mode
executor_backend = None if async_grammar else "external_launcher"
model_config = ModelConfig(tokenizer=TOKENIZER)
vllm_config = VllmConfig(
model_config=model_config,
renderer_config=RendererConfig(model_config=model_config, tokenizer=TOKENIZER),
model_config=ModelConfig(tokenizer=TOKENIZER),
structured_outputs_config=StructuredOutputsConfig(backend="guidance"),
parallel_config=ParallelConfig(distributed_executor_backend=executor_backend),
)
@@ -7,7 +7,7 @@ from unittest.mock import Mock
import pytest
from vllm.config import ModelConfig, RendererConfig, SchedulerConfig, VllmConfig
from vllm.config import ModelConfig, SchedulerConfig, VllmConfig
from vllm.reasoning import ReasoningParser
from vllm.v1.request import Request
from vllm.v1.structured_output import StructuredOutputManager
@@ -17,26 +17,19 @@ class TestReasoningStructuredOutput:
"""Test reasoning-aware structured output functionality."""
@pytest.fixture
def mock_renderer_config(self):
"""Create a mock RendererConfig."""
renderer_config = Mock(spec=RendererConfig)
renderer_config.skip_tokenizer_init = (
True # Skip tokenizer init to avoid network calls
)
model_config = Mock(spec=ModelConfig)
model_config.get_vocab_size = Mock(return_value=50000)
model_config.trust_remote_code = False
def mock_model_config(self):
"""Create a mock ModelConfig."""
config = Mock(spec=ModelConfig)
config.skip_tokenizer_init = True # Skip tokenizer init to avoid network calls
config.get_vocab_size = Mock(return_value=50000)
# Add missing runner_type attribute that tokenizer initialization expects
model_config.runner_type = "generate"
renderer_config.model_config = model_config
config.runner_type = "generate"
# Add other attributes that tokenizer initialization might need
renderer_config.tokenizer = "test-tokenizer"
renderer_config.tokenizer_mode = "auto"
renderer_config.tokenizer_revision = None
return renderer_config
config.tokenizer = "test-tokenizer"
config.tokenizer_mode = "auto"
config.trust_remote_code = False
config.tokenizer_revision = None
return config
@pytest.fixture
def mock_scheduler_config(self):
@@ -46,10 +39,10 @@ class TestReasoningStructuredOutput:
return config
@pytest.fixture
def mock_vllm_config(self, mock_renderer_config, mock_scheduler_config):
def mock_vllm_config(self, mock_model_config, mock_scheduler_config):
"""Create a mock VllmConfig."""
config = Mock(spec=VllmConfig)
config.renderer_config = mock_renderer_config
config.model_config = mock_model_config
config.scheduler_config = mock_scheduler_config
config.structured_outputs_config = Mock()
config.structured_outputs_config.reasoning_parser = None
@@ -7,7 +7,6 @@ from vllm.attention.layer import Attention
from vllm.config import (
CacheConfig,
ModelConfig,
RendererConfig,
SchedulerConfig,
VllmConfig,
set_current_vllm_config,
@@ -46,7 +45,6 @@ def get_vllm_config():
)
vllm_config = VllmConfig(
model_config=model_config,
renderer_config=RendererConfig(model_config=model_config),
cache_config=cache_config,
scheduler_config=scheduler_config,
)
-3
View File
@@ -13,7 +13,6 @@ from vllm.config import (
CacheConfig,
ModelConfig,
ParallelConfig,
RendererConfig,
SchedulerConfig,
VllmConfig,
set_current_vllm_config,
@@ -102,7 +101,6 @@ def get_vllm_config():
parallel_config = ParallelConfig()
vllm_config = VllmConfig(
model_config=model_config,
renderer_config=RendererConfig(model_config=model_config),
cache_config=cache_config,
scheduler_config=scheduler_config,
parallel_config=parallel_config,
@@ -813,7 +811,6 @@ def test_hybrid_attention_mamba_tensor_shapes():
attention_config = AttentionConfig(backend=AttentionBackendEnum.FLASHINFER)
vllm_config = VllmConfig(
model_config=model_config,
renderer_config=RendererConfig(model_config=model_config),
cache_config=cache_config,
scheduler_config=scheduler_config,
parallel_config=parallel_config,