[Feature] Use paddle.compat.enable_torch_proxy in fastdeploy/__init__.py (#5211)

* test feature

* fix xgrammar

* fix paddleformer

* try whitelist

* manual patch PaddlePaddle/Paddle#76706 for test

* remove triton version

* add comment

* Update scripts/run_ci_xpu.sh

Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>

* remove use-triton-in-paddle in requirement.txt

---------

Co-authored-by: Nyakku Shigure <sigure.qaq@gmail.com>
Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>

fastdeploy/__init__.py

@@ -31,6 +31,8 @@ if os.getenv("PROMETHEUS_MULTIPROC_DIR", "") == "":
 
 import typing
 
+import paddle
+
 # first import prometheus setup to set PROMETHEUS_MULTIPROC_DIR
 # otherwise, the Prometheus package will be imported first,
 # which will prevent correct multi-process setup
@@ -40,12 +42,24 @@ from fastdeploy.metrics.prometheus_multiprocess_setup import (
 
 setup_multiprocess_prometheus()
 
+
 from paddleformers.utils.log import logger as pf_logger
 
 from fastdeploy.engine.sampling_params import SamplingParams
 from fastdeploy.entrypoints.llm import LLM
 from fastdeploy.utils import current_package_version, envs
 
+paddle.compat.enable_torch_proxy(scope={"triton"})
+# paddle.compat.enable_torch_proxy(scope={"triton"}) enables the torch proxy
+# specifically for the 'triton' module. This means `import torch` inside 'triton'
+# will actually import paddle's compatibility layer (acting as torch).
+#
+# 'scope' acts as an allowlist. To add other modules, you can do:
+# paddle.compat.enable_torch_proxy(scope={"triton", "new_module"})
+#
+# Note: Ensure that any torch APIs used in 'new_module' are already implemented in Paddle.
+
+
 if envs.FD_DEBUG != 1:
     import logging
 

requirements.txt

@@ -26,7 +26,6 @@ prometheus-client
 decord
 moviepy
 triton
-use-triton-in-paddle
 crcmod
 msgpack
 gunicorn