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

Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk>
This commit is contained in:
Cyrus Leung
2025-12-06 23:15:42 -08:00
committed by GitHub
parent a49d813fa8
commit 27f4c2fd46
105 changed files with 969 additions and 797 deletions
@@ -17,6 +17,7 @@ from vllm.config import (
DeviceConfig,
ModelConfig,
PassConfig,
RendererConfig,
VllmConfig,
get_current_vllm_config,
set_current_vllm_config,
@@ -276,6 +277,7 @@ def sequence_parallelism_pass_on_test_model(
vllm_config = VllmConfig(
model_config=model_config,
renderer_config=RendererConfig(model_config=model_config),
device_config=device_config,
compilation_config=compilation_config,
)
+5 -1
View File
@@ -15,6 +15,7 @@ from vllm.config import (
CompilationConfig,
ModelConfig,
PassConfig,
RendererConfig,
VllmConfig,
set_current_vllm_config,
)
@@ -219,8 +220,11 @@ def test_fix_functionalization(
torch.set_default_device("cuda")
torch.set_default_dtype(dtype)
model_config = ModelConfig(dtype=dtype)
vllm_config = VllmConfig(
model_config=ModelConfig(dtype=dtype),
model_config=model_config,
renderer_config=RendererConfig(model_config=model_config),
compilation_config=CompilationConfig(
custom_ops=["all"],
pass_config=PassConfig(
+5 -1
View File
@@ -15,6 +15,7 @@ from vllm.config import (
CompilationMode,
ModelConfig,
PassConfig,
RendererConfig,
VllmConfig,
)
from vllm.model_executor.layers.layernorm import RMSNorm
@@ -154,8 +155,11 @@ def test_fusion_rmsnorm_quant(
custom_ops.append("+rms_norm")
if enable_quant_fp8_custom_op:
custom_ops.append("+quant_fp8")
model_config = ModelConfig(dtype=dtype)
vllm_config = VllmConfig(
model_config=ModelConfig(dtype=dtype),
model_config=model_config,
renderer_config=RendererConfig(model_config=model_config),
compilation_config=CompilationConfig(
mode=CompilationMode.VLLM_COMPILE,
custom_ops=custom_ops,
+2
View File
@@ -24,6 +24,7 @@ from vllm.config import (
CompilationMode,
ModelConfig,
PassConfig,
RendererConfig,
SchedulerConfig,
VllmConfig,
set_current_vllm_config,
@@ -325,6 +326,7 @@ def test_attention_quant_pattern(
)
vllm_config = VllmConfig(
model_config=model_config,
renderer_config=RendererConfig(model_config=model_config),
scheduler_config=SchedulerConfig(
max_num_seqs=1024,
max_model_len=model_config.max_model_len,
+6 -2
View File
@@ -7,7 +7,7 @@ import torch
from vllm.compilation.inductor_pass import CallableInductorPass, InductorPass
from vllm.compilation.pass_manager import PostGradPassManager
from vllm.config import ModelConfig, VllmConfig
from vllm.config import ModelConfig, RendererConfig, VllmConfig
# dummy custom pass that doesn't inherit
@@ -43,7 +43,11 @@ class ProperPass(InductorPass):
)
def test_pass_manager_uuid(callable):
# Some passes need dtype to be set
config = VllmConfig(model_config=ModelConfig(dtype=torch.bfloat16))
model_config = ModelConfig(dtype=torch.bfloat16)
config = VllmConfig(
model_config=model_config,
renderer_config=RendererConfig(model_config=model_config),
)
pass_manager = PostGradPassManager()
pass_manager.configure(config)
+4 -1
View File
@@ -19,6 +19,7 @@ from vllm.config import (
CompilationMode,
ModelConfig,
PassConfig,
RendererConfig,
VllmConfig,
set_current_vllm_config,
)
@@ -133,8 +134,10 @@ def test_qk_norm_rope_fusion(
if enable_rope_custom_op:
custom_ops.append("+rotary_embedding")
model_config = ModelConfig(dtype=dtype)
vllm_config = VllmConfig(
model_config=ModelConfig(dtype=dtype),
model_config=model_config,
renderer_config=RendererConfig(model_config=model_config),
compilation_config=CompilationConfig(
mode=CompilationMode.VLLM_COMPILE,
custom_ops=custom_ops,