diff --git a/.github/workflows/pr-test.yml b/.github/workflows/pr-test.yml
index 5ac065973..3b33b319d 100644
--- a/.github/workflows/pr-test.yml
+++ b/.github/workflows/pr-test.yml
@@ -95,7 +95,7 @@ jobs:
     strategy:
       fail-fast: false
       matrix:
-        range: [0-6, 6-15, 15-22, 22-32, 32-40, 40-48, 48-100]
+        part: [0, 1, 2, 3, 4, 5, 6]
     steps:
       - name: Checkout code
         uses: actions/checkout@v3
@@ -109,11 +109,8 @@ jobs:
       - name: Run test
         timeout-minutes: 30
         run: |
-          RANGE=${{ matrix.range }}
-          range_begin=${RANGE%-*}
-          range_end=${RANGE#*-}
           cd test/srt
-          python3 run_suite.py --suite per-commit --range-begin ${range_begin} --range-end ${range_end}
+          python3 run_suite.py --suite per-commit --auto-partition-id ${{ matrix.part }} --auto-partition-size 7
 
   unit-test-backend-2-gpu:
     needs: filter
@@ -340,7 +337,6 @@ jobs:
           python3 test_moe_eval_accuracy_large.py
 
   finish:
-    if: always()
     needs: [
       unit-test-frontend, unit-test-backend-1-gpu, unit-test-backend-2-gpu,
       performance-test-1-gpu-part-1, performance-test-1-gpu-part-2, performance-test-2-gpu,
diff --git a/python/sglang/test/test_utils.py b/python/sglang/test/test_utils.py
index a8bf674a9..e2d92bb51 100644
--- a/python/sglang/test/test_utils.py
+++ b/python/sglang/test/test_utils.py
@@ -446,22 +446,31 @@ def run_with_timeout(
     return ret_value[0]
 
 
-def run_unittest_files(files: List[str], timeout_per_file: float):
+def run_unittest_files(files: List, timeout_per_file: float):
     tic = time.time()
     success = True
 
-    for filename in files:
+    for file in files:
+        filename, estimated_time = file.name, file.estimated_time
         process = None
 
         def run_one_file(filename):
             nonlocal process
 
             filename = os.path.join(os.getcwd(), filename)
-            print(f"\n\nRun:\npython3 {filename}\n\n", flush=True)
+            print(f".\n.\nBegin:\npython3 {filename}\n.\n.\n", flush=True)
+            tic = time.time()
+
             process = subprocess.Popen(
                 ["python3", filename], stdout=None, stderr=None, env=os.environ
             )
             process.wait()
+            elapsed = time.time() - tic
+
+            print(
+                f".\n.\nEnd:\n{filename=}, {elapsed=:.0f}, {estimated_time=}\n.\n.\n",
+                flush=True,
+            )
             return process.returncode
 
         try:
diff --git a/test/lang/run_suite.py b/test/lang/run_suite.py
index 327d18b3f..e4e11ca4f 100644
--- a/test/lang/run_suite.py
+++ b/test/lang/run_suite.py
@@ -1,11 +1,19 @@
 import argparse
 import glob
+from dataclasses import dataclass
 
 from sglang.test.test_utils import run_unittest_files
 
+
+@dataclass
+class TestFile:
+    name: str
+    estimated_time: float = 60
+
+
 suites = {
     "per-commit": [
-        "test_srt_backend.py",
+        TestFile("test_srt_backend.py"),
         # Skip this due to some OPENAI_API_KEY issues
         # "test_openai_backend.py",
     ],
diff --git a/test/srt/run_suite.py b/test/srt/run_suite.py
index ebab2bf68..b9e36e232 100644
--- a/test/srt/run_suite.py
+++ b/test/srt/run_suite.py
@@ -1,85 +1,125 @@
 import argparse
 import glob
+from dataclasses import dataclass
 
 from sglang.test.test_utils import run_unittest_files
 
+
+@dataclass
+class TestFile:
+    name: str
+    estimated_time: float = 60
+
+
 suites = {
     "per-commit": [
-        "models/lora/test_lora.py",
-        "models/lora/test_lora_backend.py",
-        "models/lora/test_multi_lora_backend.py",
-        "models/test_embedding_models.py",
-        "models/test_generation_models.py",
-        "models/test_qwen_models.py",
-        "models/test_reward_models.py",
-        "test_gptqmodel_dynamic.py",
-        "models/test_gme_qwen_models.py",
-        "test_abort.py",
-        "test_chunked_prefill.py",
-        "test_custom_allreduce.py",
-        "test_double_sparsity.py",
-        "test_eagle_infer.py",
-        "test_embedding_openai_server.py",
-        "test_eval_accuracy_mini.py",
-        "test_gguf.py",
-        "test_input_embeddings.py",
-        "test_mla.py",
-        "test_mla_deepseek_v3.py",
-        "test_mla_flashinfer.py",
-        "test_mla_fp8.py",
-        "test_json_constrained.py",
-        "test_large_max_new_tokens.py",
-        "test_metrics.py",
-        "test_no_chunked_prefill.py",
-        "test_no_overlap_scheduler.py",
-        "test_openai_server.py",
-        "test_penalty.py",
-        "test_pytorch_sampling_backend.py",
-        "test_radix_attention.py",
-        "test_regex_constrained.py",
-        "test_release_memory_occupation.py",
-        "test_request_length_validation.py",
-        "test_retract_decode.py",
-        "test_server_args.py",
-        # Disabled temporarily
-        # "test_session_control.py",
-        "test_skip_tokenizer_init.py",
-        "test_srt_engine.py",
-        "test_srt_endpoint.py",
-        "test_torch_compile.py",
-        "test_torch_compile_moe.py",
-        "test_torch_native_attention_backend.py",
-        "test_torchao.py",
-        "test_triton_attention_kernels.py",
-        "test_triton_attention_backend.py",
-        "test_hidden_states.py",
-        "test_update_weights_from_disk.py",
-        "test_update_weights_from_tensor.py",
-        "test_vertex_endpoint.py",
-        "test_vision_chunked_prefill.py",
-        "test_vision_llm.py",
-        "test_vision_openai_server.py",
-        "test_w8a8_quantization.py",
-        "test_fp8_kernel.py",
-        "test_block_int8.py",
-        "test_int8_kernel.py",
-        "test_reasoning_content.py",
+        TestFile("models/lora/test_lora.py", 76),
+        TestFile("models/lora/test_lora_backend.py", 420),
+        TestFile("models/lora/test_multi_lora_backend.py", 1),
+        TestFile("models/test_embedding_models.py", 119),
+        TestFile("models/test_generation_models.py", 103),
+        TestFile("models/test_qwen_models.py", 82),
+        TestFile("models/test_reward_models.py", 83),
+        TestFile("test_gptqmodel_dynamic.py", 72),
+        TestFile("models/test_gme_qwen_models.py", 45),
+        TestFile("test_abort.py", 51),
+        TestFile("test_chunked_prefill.py", 336),
+        TestFile("test_custom_allreduce.py", 1),
+        TestFile("test_double_sparsity.py", 50),
+        TestFile("test_eagle_infer.py", 447),
+        TestFile("test_embedding_openai_server.py", 36),
+        TestFile("test_eval_accuracy_mini.py", 63),
+        TestFile("test_gguf.py", 78),
+        TestFile("test_input_embeddings.py", 38),
+        TestFile("test_mla.py", 92),
+        TestFile("test_mla_deepseek_v3.py", 221),
+        TestFile("test_mla_flashinfer.py", 395),
+        TestFile("test_mla_fp8.py", 93),
+        TestFile("test_json_constrained.py", 98),
+        TestFile("test_large_max_new_tokens.py", 41),
+        TestFile("test_metrics.py", 32),
+        TestFile("test_no_chunked_prefill.py", 126),
+        TestFile("test_no_overlap_scheduler.py", 262),
+        TestFile("test_openai_server.py", 124),
+        TestFile("test_penalty.py", 41),
+        TestFile("test_pytorch_sampling_backend.py", 66),
+        TestFile("test_radix_attention.py", 167),
+        TestFile("test_regex_constrained.py", 64),
+        TestFile("test_release_memory_occupation.py", 44),
+        TestFile("test_request_length_validation.py", 31),
+        TestFile("test_retract_decode.py", 54),
+        TestFile("test_server_args.py", 1),
+        TestFile("test_skip_tokenizer_init.py", 72),
+        TestFile("test_srt_engine.py", 237),
+        TestFile("test_srt_endpoint.py", 94),
+        TestFile("test_torch_compile.py", 76),
+        TestFile("test_torch_compile_moe.py", 85),
+        TestFile("test_torch_native_attention_backend.py", 149),
+        TestFile("test_torchao.py", 70),
+        TestFile("test_triton_attention_kernels.py", 4),
+        TestFile("test_triton_attention_backend.py", 134),
+        TestFile("test_hidden_states.py", 55),
+        TestFile("test_update_weights_from_disk.py", 114),
+        TestFile("test_update_weights_from_tensor.py", 48),
+        TestFile("test_vertex_endpoint.py", 31),
+        TestFile("test_vision_chunked_prefill.py", 223),
+        TestFile("test_vision_llm.py", 18.4),
+        TestFile("test_vision_openai_server.py", 344),
+        TestFile("test_w8a8_quantization.py", 46),
+        TestFile("test_fp8_kernel.py", 2),
+        TestFile("test_block_int8.py", 22),
+        TestFile("test_int8_kernel.py", 1),
+        TestFile("test_reasoning_content.py", 89),
     ],
     "nightly": [
-        "test_nightly_gsm8k_eval.py",
-        # Disable temporarily
-        # "test_nightly_math_eval.py",
+        TestFile("test_nightly_gsm8k_eval.py"),
     ],
 }
 
-# Expand suite
-for target_suite_name, target_tests in suites.items():
-    for suite_name, tests in suites.items():
-        if suite_name == target_suite_name:
-            continue
-        if target_suite_name in tests:
-            tests.remove(target_suite_name)
-            tests.extend(target_tests)
+
+def auto_partition(files, rank, size):
+    """
+    Partition files into size sublists with approximately equal sums of estimated times
+    using stable sorting, and return the partition for the specified rank.
+
+    Args:
+        files (list): List of file objects with estimated_time attribute
+        rank (int): Index of the partition to return (0 to size-1)
+        size (int): Number of partitions
+
+    Returns:
+        list: List of file objects in the specified rank's partition
+    """
+    weights = [f.estimated_time for f in files]
+
+    if not weights or size <= 0 or size > len(weights):
+        return []
+
+    # Create list of (weight, original_index) tuples
+    # Using negative index as secondary key to maintain original order for equal weights
+    indexed_weights = [(w, -i) for i, w in enumerate(weights)]
+    # Stable sort in descending order by weight
+    # If weights are equal, larger (negative) index comes first (i.e., earlier original position)
+    indexed_weights = sorted(indexed_weights, reverse=True)
+
+    # Extract original indices (negate back to positive)
+    indexed_weights = [(w, -i) for w, i in indexed_weights]
+
+    # Initialize partitions and their sums
+    partitions = [[] for _ in range(size)]
+    sums = [0.0] * size
+
+    # Greedy approach: assign each weight to partition with smallest current sum
+    for weight, idx in indexed_weights:
+        # Find partition with minimum sum
+        min_sum_idx = sums.index(min(sums))
+        partitions[min_sum_idx].append(idx)
+        sums[min_sum_idx] += weight
+
+    # Return the files corresponding to the indices in the specified rank's partition
+    indices = partitions[rank]
+    return [files[i] for i in indices]
+
 
 if __name__ == "__main__":
     arg_parser = argparse.ArgumentParser()
@@ -108,17 +148,30 @@ if __name__ == "__main__":
         default=None,
         help="The end index of the range of the files to run.",
     )
+    arg_parser.add_argument(
+        "--auto-partition-id",
+        type=int,
+        help="Use auto load balancing. The part id.",
+    )
+    arg_parser.add_argument(
+        "--auto-partition-size",
+        type=int,
+        help="Use auto load balancing. The number of parts.",
+    )
     args = arg_parser.parse_args()
+    print(f"{args=}")
 
     if args.suite == "all":
         files = glob.glob("**/test_*.py", recursive=True)
     else:
         files = suites[args.suite]
 
-    files = files[args.range_begin : args.range_end]
+    if args.auto_partition_size:
+        files = auto_partition(files, args.auto_partition_id, args.auto_partition_size)
+    else:
+        files = files[args.range_begin : args.range_end]
 
-    print(f"{args=}")
-    print("The running tests are ", files)
+    print("The running tests are ", [f.name for f in files])
 
     exit_code = run_unittest_files(files, args.timeout_per_file)
     exit(exit_code)
diff --git a/test/srt/test_custom_allreduce.py b/test/srt/test_custom_allreduce.py
index 5f6f5d9b4..7ac0c8ffc 100644
--- a/test/srt/test_custom_allreduce.py
+++ b/test/srt/test_custom_allreduce.py
@@ -42,7 +42,8 @@ def multi_process_parallel(
     # as compared to multiprocessing.
     # NOTE: We need to set working_dir for distributed tests,
     # otherwise we may get import errors on ray workers
-    ray.init(log_to_driver=False)
+
+    ray.init(log_to_driver=True)
 
     distributed_init_port = get_open_port()
     refs = []