[diffusion]: align sglang diffusion AMD pyproject_other.toml diffusion dependency with pyproject.toml (#16225)

Co-authored-by: roywang <roywang@amd.com>

docker/rocm.Dockerfile

@@ -183,7 +183,7 @@ RUN git clone ${SGL_REPO} \
     && if [ "$BUILD_TYPE" = "srt" ]; then \
          python -m pip --no-cache-dir install -e "python[srt_hip,diffusion_hip]"; \
        else \
-         python -m pip --no-cache-dir install -e "python[all_hip,diffusion_hip]"; \
+         python -m pip --no-cache-dir install -e "python[all_hip]"; \
        fi
 
 RUN python -m pip cache purge

docs/platforms/amd_gpu.md

@@ -55,7 +55,7 @@ python setup_rocm.py install
 # Install sglang python package along with diffusion support
 cd ..
 rm -rf python/pyproject.toml && mv python/pyproject_other.toml python/pyproject.toml
-pip install -e "python[all_hip,diffusion_hip]"
+pip install -e "python[all_hip]"
 ```
 
 ### Install Using Docker (Recommended)

python/pyproject_other.toml

@@ -85,18 +85,18 @@ srt_hip = [
 ]
 
 diffusion_hip = [
-  "diffusers @ git+https://github.com/huggingface/diffusers.git@6290fdfda40610ce7b99920146853614ba529c6e",
-  "opencv-python-headless==4.10.0.84",
+  "PyYAML==6.0.1",
+  "cloudpickle",
+  "diffusers==0.36.0",
   "imageio==2.36.0",
   "imageio-ffmpeg==0.5.1",
-  "PyYAML==6.0.1",
   "moviepy>=2.0.0",
-  "cloudpickle",
+  "opencv-python-headless==4.10.0.84",
   "remote-pdb",
-  "torchcodec==0.5.0",
   "st_attn==0.0.7",
   "vsa==0.0.4",
   "runai_model_streamer>=0.15.5",
+  "cache-dit==1.1.8"
 ]
 
 # For Intel Gaudi(device : hpu) follow the installation guide
@@ -141,7 +141,7 @@ test = [
   "tabulate",
 ]
 
-all_hip = ["sglang[srt_hip]"]
+all_hip = ["sglang[srt_hip]", "sglang[diffusion_hip]"]
 all_hpu = ["sglang[srt_hpu]"]
 all_musa = ["sglang[srt_musa]", "sglang[diffusion_musa]"]
 

python/sglang/multimodal_gen/runtime/loader/weight_utils.py

@@ -23,11 +23,6 @@ try:
 except ImportError:
     HAS_RUNAI_MODEL_STREAMER = False
 
-# Disable runai_model_streamer on AMD/ROCm due to global state issues
-# that cause "Streamer is handling previous request" errors
-if torch.version.hip is not None:
-    HAS_RUNAI_MODEL_STREAMER = False
-
 from sglang.multimodal_gen.runtime.distributed import get_local_torch_device
 from sglang.multimodal_gen.runtime.utils.logging_utils import init_logger