[diffusion] fix: replace TextEncoderConfig with Qwen3TextConfig for Z-Image (#18560)
Co-authored-by: Mick <mickjagger19@icloud.com>
This commit is contained in:
@@ -7,10 +7,8 @@ import torch
|
|||||||
|
|
||||||
from sglang.multimodal_gen.configs.models import DiTConfig, EncoderConfig, VAEConfig
|
from sglang.multimodal_gen.configs.models import DiTConfig, EncoderConfig, VAEConfig
|
||||||
from sglang.multimodal_gen.configs.models.dits.zimage import ZImageDitConfig
|
from sglang.multimodal_gen.configs.models.dits.zimage import ZImageDitConfig
|
||||||
from sglang.multimodal_gen.configs.models.encoders import (
|
from sglang.multimodal_gen.configs.models.encoders import BaseEncoderOutput
|
||||||
BaseEncoderOutput,
|
from sglang.multimodal_gen.configs.models.encoders.qwen3 import Qwen3TextConfig
|
||||||
TextEncoderConfig,
|
|
||||||
)
|
|
||||||
from sglang.multimodal_gen.configs.models.vaes.flux import FluxVAEConfig
|
from sglang.multimodal_gen.configs.models.vaes.flux import FluxVAEConfig
|
||||||
from sglang.multimodal_gen.configs.pipeline_configs.base import (
|
from sglang.multimodal_gen.configs.pipeline_configs.base import (
|
||||||
ImagePipelineConfig,
|
ImagePipelineConfig,
|
||||||
@@ -49,7 +47,7 @@ class ZImagePipelineConfig(ImagePipelineConfig):
|
|||||||
dit_config: DiTConfig = field(default_factory=ZImageDitConfig)
|
dit_config: DiTConfig = field(default_factory=ZImageDitConfig)
|
||||||
vae_config: VAEConfig = field(default_factory=FluxVAEConfig)
|
vae_config: VAEConfig = field(default_factory=FluxVAEConfig)
|
||||||
text_encoder_configs: tuple[EncoderConfig, ...] = field(
|
text_encoder_configs: tuple[EncoderConfig, ...] = field(
|
||||||
default_factory=lambda: (TextEncoderConfig(),)
|
default_factory=lambda: (Qwen3TextConfig(),)
|
||||||
)
|
)
|
||||||
|
|
||||||
preprocess_text_funcs: tuple[Callable, ...] = field(
|
preprocess_text_funcs: tuple[Callable, ...] = field(
|
||||||
|
|||||||
@@ -525,7 +525,7 @@
|
|||||||
"stages_ms": {
|
"stages_ms": {
|
||||||
"ConditioningStage": 0.01,
|
"ConditioningStage": 0.01,
|
||||||
"DecodingStage": 33.93,
|
"DecodingStage": 33.93,
|
||||||
"TextEncodingStage": 104.21,
|
"TextEncodingStage": 409.74,
|
||||||
"InputValidationStage": 0.03,
|
"InputValidationStage": 0.03,
|
||||||
"DenoisingStage": 808.69,
|
"DenoisingStage": 808.69,
|
||||||
"LatentPreparationStage": 0.14,
|
"LatentPreparationStage": 0.14,
|
||||||
@@ -542,14 +542,14 @@
|
|||||||
"7": 99.3,
|
"7": 99.3,
|
||||||
"8": 100.83
|
"8": 100.83
|
||||||
},
|
},
|
||||||
"expected_e2e_ms": 952.09,
|
"expected_e2e_ms": 1461.08,
|
||||||
"expected_avg_denoise_ms": 89.36,
|
"expected_avg_denoise_ms": 89.36,
|
||||||
"expected_median_denoise_ms": 99.67
|
"expected_median_denoise_ms": 99.67
|
||||||
},
|
},
|
||||||
"zimage_image_t2i_multi_lora": {
|
"zimage_image_t2i_multi_lora": {
|
||||||
"stages_ms": {
|
"stages_ms": {
|
||||||
"InputValidationStage": 0.04,
|
"InputValidationStage": 0.04,
|
||||||
"TextEncodingStage": 103.81,
|
"TextEncodingStage": 413.69,
|
||||||
"ConditioningStage": 0.01,
|
"ConditioningStage": 0.01,
|
||||||
"TimestepPreparationStage": 1.3,
|
"TimestepPreparationStage": 1.3,
|
||||||
"LatentPreparationStage": 0.11,
|
"LatentPreparationStage": 0.11,
|
||||||
@@ -567,14 +567,14 @@
|
|||||||
"7": 100.86,
|
"7": 100.86,
|
||||||
"8": 103.87
|
"8": 103.87
|
||||||
},
|
},
|
||||||
"expected_e2e_ms": 955.14,
|
"expected_e2e_ms": 1464.31,
|
||||||
"expected_avg_denoise_ms": 89.96,
|
"expected_avg_denoise_ms": 89.96,
|
||||||
"expected_median_denoise_ms": 99.72
|
"expected_median_denoise_ms": 99.72
|
||||||
},
|
},
|
||||||
"zimage_image_t2i_2_gpus": {
|
"zimage_image_t2i_2_gpus": {
|
||||||
"stages_ms": {
|
"stages_ms": {
|
||||||
"InputValidationStage": 0.08,
|
"InputValidationStage": 0.08,
|
||||||
"TextEncodingStage": 118.66,
|
"TextEncodingStage": 420.74,
|
||||||
"ConditioningStage": 0.01,
|
"ConditioningStage": 0.01,
|
||||||
"TimestepPreparationStage": 1.5,
|
"TimestepPreparationStage": 1.5,
|
||||||
"LatentPreparationStage": 0.12,
|
"LatentPreparationStage": 0.12,
|
||||||
@@ -599,7 +599,7 @@
|
|||||||
"zimage_image_t2i_warmup": {
|
"zimage_image_t2i_warmup": {
|
||||||
"stages_ms": {
|
"stages_ms": {
|
||||||
"InputValidationStage": 0.02,
|
"InputValidationStage": 0.02,
|
||||||
"TextEncodingStage": 100.65,
|
"TextEncodingStage": 456.16,
|
||||||
"ConditioningStage": 0.01,
|
"ConditioningStage": 0.01,
|
||||||
"TimestepPreparationStage": 0.98,
|
"TimestepPreparationStage": 0.98,
|
||||||
"LatentPreparationStage": 0.06,
|
"LatentPreparationStage": 0.06,
|
||||||
@@ -617,7 +617,7 @@
|
|||||||
"7": 110.54,
|
"7": 110.54,
|
||||||
"8": 115.24
|
"8": 115.24
|
||||||
},
|
},
|
||||||
"expected_e2e_ms": 1029.96,
|
"expected_e2e_ms": 1184.45,
|
||||||
"expected_avg_denoise_ms": 98.46,
|
"expected_avg_denoise_ms": 98.46,
|
||||||
"expected_median_denoise_ms": 109.65
|
"expected_median_denoise_ms": 109.65
|
||||||
},
|
},
|
||||||
@@ -2001,7 +2001,7 @@
|
|||||||
"fsdp-inference": {
|
"fsdp-inference": {
|
||||||
"stages_ms": {
|
"stages_ms": {
|
||||||
"InputValidationStage": 0.04,
|
"InputValidationStage": 0.04,
|
||||||
"TextEncodingStage": 128.3,
|
"TextEncodingStage": 411.12,
|
||||||
"ConditioningStage": 0.01,
|
"ConditioningStage": 0.01,
|
||||||
"TimestepPreparationStage": 1.44,
|
"TimestepPreparationStage": 1.44,
|
||||||
"LatentPreparationStage": 0.1,
|
"LatentPreparationStage": 0.1,
|
||||||
@@ -2019,7 +2019,7 @@
|
|||||||
"7": 178.53,
|
"7": 178.53,
|
||||||
"8": 178.08
|
"8": 178.08
|
||||||
},
|
},
|
||||||
"expected_e2e_ms": 1742.7,
|
"expected_e2e_ms": 2103.05,
|
||||||
"expected_avg_denoise_ms": 173.83,
|
"expected_avg_denoise_ms": 173.83,
|
||||||
"expected_median_denoise_ms": 178.08
|
"expected_median_denoise_ms": 178.08
|
||||||
}
|
}
|
||||||
|
|||||||
Reference in New Issue
Block a user