From e319153be8c5903d2b652d50725b308ec2e99a40 Mon Sep 17 00:00:00 2001
From: Yineng Zhang <me@zhyncs.com>
Date: Mon, 17 Feb 2025 21:06:10 +0800
Subject: [PATCH] update unit test (#3636)

---
 .github/workflows/pr-test.yml | 2 ++
 test/srt/run_suite.py         | 4 ++--
 test/srt/test_mla.py          | 7 +++----
 3 files changed, 7 insertions(+), 6 deletions(-)

diff --git a/.github/workflows/pr-test.yml b/.github/workflows/pr-test.yml
index 69aa1ee7b..510a741e7 100644
--- a/.github/workflows/pr-test.yml
+++ b/.github/workflows/pr-test.yml
@@ -107,6 +107,8 @@ jobs:
           bash scripts/ci_install_dependency.sh
 
       - name: Run test
+        env:
+          HF_TOKEN: ${{ secrets.HF_TOKEN }}
         timeout-minutes: 30
         run: |
           RANGE=${{ matrix.range }}
diff --git a/test/srt/run_suite.py b/test/srt/run_suite.py
index d263bc113..d2323d426 100644
--- a/test/srt/run_suite.py
+++ b/test/srt/run_suite.py
@@ -21,11 +21,11 @@ suites = {
         "test_eval_accuracy_mini.py",
         "test_gguf.py",
         "test_input_embeddings.py",
+        "test_mla.py",
+        "test_mla_fp8.py",
         "test_json_constrained.py",
         "test_large_max_new_tokens.py",
         "test_metrics.py",
-        "test_mla.py",
-        "test_mla_fp8.py",
         "test_no_chunked_prefill.py",
         "test_no_overlap_scheduler.py",
         "test_openai_server.py",
diff --git a/test/srt/test_mla.py b/test/srt/test_mla.py
index 87351c184..c7016eb14 100644
--- a/test/srt/test_mla.py
+++ b/test/srt/test_mla.py
@@ -58,7 +58,7 @@ class TestMLA(unittest.TestCase):
 class TestDeepseekV3(unittest.TestCase):
     @classmethod
     def setUpClass(cls):
-        cls.model = "lmzheng/sglang-ci-dsv3-test"
+        cls.model = "sgl-project/sglang-ci-dsv3-test"
         cls.base_url = DEFAULT_URL_FOR_TEST
         other_args = ["--trust-remote-code"]
         if torch.cuda.is_available() and torch.version.cuda:
@@ -93,13 +93,12 @@ class TestDeepseekV3(unittest.TestCase):
 class TestDeepseekV3MTP(unittest.TestCase):
     @classmethod
     def setUpClass(cls):
-        cls.model = "lmzheng/sglang-ci-dsv3-test"
+        cls.model = "sgl-project/sglang-ci-dsv3-test"
         cls.base_url = DEFAULT_URL_FOR_TEST
         other_args = ["--trust-remote-code"]
         if torch.cuda.is_available() and torch.version.cuda:
             other_args.extend(
                 [
-                    "--enable-torch-compile",
                     "--cuda-graph-max-bs",
                     "2",
                     "--disable-radix",
@@ -109,7 +108,7 @@ class TestDeepseekV3MTP(unittest.TestCase):
                     "--speculative-algorithm",
                     "NEXTN",
                     "--speculative-draft",
-                    "SGLang/sglang-ci-dsv3-test-NextN",
+                    "sgl-project/sglang-ci-dsv3-test-NextN",
                     "--speculative-num-steps",
                     "2",
                     "--speculative-eagle-topk",