fix issues

Author: billishyahao

setup.py

@@ -159,21 +159,29 @@ def run(self):
         os.chdir('thirdparties')
 
         # flash-attention
+        # MI3xx has both CK backend and Triton backend.
         if (rocm_arch := os.environ['ROCM_ARCH']) in INSTINCT_ARCH:
             subprocess.check_call(['git', 'clone', '--recursive', 'https://github.com/ROCm/flash-attention.git'])
             os.chdir('flash-attention')
             num_jobs = os.cpu_count() - 1
             subprocess.check_call(['pip', 'install', '-v', '.', f'MAX_JOBS={num_jobs}'], shell=True)
             os.chdir('..')
+        # Only Triton backend supports Radeon GPUs
+        elif (rocm_arch := os.environ['ROCM_ARCH']) in RADEON_ARCH:
+            subprocess.check_call(['git', 'clone', '--recursive', 'https://github.com/ROCm/flash-attention.git'])
+            os.chdir('flash-attention')
+            subprocess.check_call(['git', 'checkout', 'main_perf'])
+            subprocess.check_call(['FLASH_ATTENTION_TRITON_AMD_ENABLE="TRUE"', 'python', 'setup.py', 'install', ], shell=True)
+            os.chdir('..')
 
-        # # xformers
-        rocm_arch = os.environ['ROCM_ARCH']
-        subprocess.check_call(['git', 'clone', 'https://github.com/ROCm/xformers.git'])
-        os.chdir('xformers')
-        subprocess.check_call(['git', 'submodule', 'update', '--init', '--recursive'])
-        os.environ['PYTORCH_ROCM_ARCH'] = rocm_arch
-        subprocess.check_call(['python', 'setup.py', 'install'])
-        os.chdir('..')
+        # only install xformers in Instinct GPUs
+        if (rocm_arch := os.environ['ROCM_ARCH']) in INSTINCT_ARCH:
+            subprocess.check_call(['git', 'clone', 'https://github.com/ROCm/xformers.git'])
+            os.chdir('xformers')
+            subprocess.check_call(['git', 'submodule', 'update', '--init', '--recursive'])
+            os.environ['PYTORCH_ROCM_ARCH'] = rocm_arch
+            subprocess.check_call(['python', 'setup.py', 'install'])
+            os.chdir('..')
 
         # bitsandbytes
         subprocess.check_call(['git', 'clone', '--recurse-submodules', 'https://github.com/ROCm/bitsandbytes'])

use_existing_torch.py

@@ -0,0 +1,22 @@
+# SPDX-License-Identifier: Apache-2.0
+# SPDX-FileCopyrightText: Copyright contributors to the vLLM project and Unsloth project
+# Copying from https://github.com/vllm-project/vllm/blob/main/use_existing_torch.py
+
+import glob
+
+requires_files = glob.glob('requirements/*.txt')
+requires_files += ["pyproject.toml"]
+for file in requires_files:
+    print(f">>> cleaning {file}")
+    with open(file) as f:
+        lines = f.readlines()
+    if "torch" in "".join(lines).lower():
+        print("removed:")
+        with open(file, 'w') as f:
+            for line in lines:
+                if 'torch' not in line.lower():
+                    f.write(line)
+                else:
+                    print(line.strip())
+    print(f"<<< done cleaning {file}")
+    print()