[diffusion] model: GLM-Image (#16894)

Co-authored-by: jianyingzhu <53300651@qq.com>
2026-01-14 02:02:03 +08:00
parent 2a7b67adff
commit a0b4ba9032
13 changed files with 1951 additions and 3 deletions
@@ -0,0 +1,58 @@
+from sglang.multimodal_gen.runtime.models.model_stages.glm_image import (
+    GlmImageBeforeDenoisingStage,
+)
+from sglang.multimodal_gen.runtime.pipelines_core import LoRAPipeline
+from sglang.multimodal_gen.runtime.pipelines_core.composed_pipeline_base import (
+    ComposedPipelineBase,
+)
+from sglang.multimodal_gen.runtime.pipelines_core.stages import (
+    DecodingStage,
+    DenoisingStage,
+)
+from sglang.multimodal_gen.runtime.server_args import ServerArgs
+from sglang.multimodal_gen.runtime.utils.logging_utils import init_logger
+
+logger = init_logger(__name__)
+
+
+class GlmImagePipeline(LoRAPipeline, ComposedPipelineBase):
+    pipeline_name = "GlmImagePipeline"
+
+    _required_config_modules = [
+        "text_encoder",
+        "tokenizer",
+        "vae",
+        "vision_language_encoder",
+        "processor",
+        "transformer",
+        "scheduler",
+    ]
+
+    def create_pipeline_stages(self, server_args: ServerArgs):
+        self.add_stage(
+            stage_name="GlmImageBeforeDenoisingStage",
+            stage=GlmImageBeforeDenoisingStage(
+                vae=self.get_module("vae"),
+                text_encoder=self.get_module("text_encoder"),
+                tokenizer=self.get_module("tokenizer"),
+                processor=self.get_module("processor"),
+                transformer=self.get_module("transformer"),
+                scheduler=self.get_module("scheduler"),
+                vision_language_encoder=self.get_module("vision_language_encoder"),
+            ),
+        )
+
+        self.add_stage(
+            stage_name="denoising_stage",
+            stage=DenoisingStage(
+                transformer=self.get_module("transformer"),
+                scheduler=self.get_module("scheduler"),
+            ),
+        )
+
+        self.add_stage(
+            stage_name="decoding_stage", stage=DecodingStage(vae=self.get_module("vae"))
+        )
+
+
+EntryClass = [GlmImagePipeline]