fix flashinfer version with sglang

ZX-ModelCloud · Jul 30, 2024 · f25d848 · f25d848
1 parent bd0765a
commit f25d848
Showing 1 changed file with 2 additions and 2 deletions.
diff --git a/tests/test_sglang.py b/tests/test_sglang.py
@@ -16,8 +16,8 @@ class TestLoadSglang(unittest.TestCase):
     @classmethod
     def setUpClass(self):
         # sglang set disable_flashinfer=True still import flashinfer
-        subprocess.check_call([sys.executable, "-m", "pip", "install", "flashinfer", "-i", f"https://flashinfer.ai/whl/cu{torch.version.cuda.replace('.', '')}/torch{'.'.join(torch.__version__.split('.')[:2])}"])
-        subprocess.check_call([sys.executable, "-m", "pip", "install", "sglang[srt]>=0.1.19"])
+        subprocess.check_call([sys.executable, "-m", "pip", "install", "flashinfer==0.1.2", "-i", f"https://flashinfer.ai/whl/cu{torch.version.cuda.replace('.', '')}/torch{'.'.join(torch.__version__.split('.')[:2])}"])
+        subprocess.check_call([sys.executable, "-m", "pip", "install", "sglang[srt]>=0.2.7"])
 
         self.MODEL_ID = "LnL-AI/TinyLlama-1.1B-Chat-v1.0-GPTQ-4bit"
         self.prompt = "The capital of France is"