Reduction of the MINI profiler UT execution time (vllm-project#52)

* Adjust the UT parameters to reduce the total time of the MINI profiler test

Signed-off-by: chzhang <chaojun.zhang@intel.com>

* Adjust the UT parameters to reduce the total time of the MINI profiler test

Signed-off-by: chzhang <chaojun.zhang@intel.com>

* Adjust the UT parameters to reduce the total time of the MINI profiler test

Signed-off-by: chzhang <chaojun.zhang@intel.com>

---------

Signed-off-by: chzhang <chaojun.zhang@intel.com>

Author: chaojun-zhang

tests/fused_moe/test_fused_moe.py

@@ -30,9 +30,9 @@ def random_partition(size_a: int, target: int):
 
 MINI_PYTEST_PARAMS = {
     "default": {
-        "m,n,k": [(1, 256, 128), (4, 512, 256), (16, 512, 512)],
-        "e": [16, 8],
-        "topk": [1, 2],
+        "m,n,k": [(1, 256, 128)],
+        "e": [2],
+        "topk": [1],
         "dtype": [torch.bfloat16]
     }
 }

tests/test_cache.py

@@ -40,15 +40,19 @@
 MINI_PYTEST_PARAMS = {
     "default": {
         "num_tokens": [1],
-        "head_size": [64, 80],
+        "head_size": [8],
+        "num_blocks": [4],
+        "block_size": [8],
     },
     "test_concat_and_cache_mla": {
         "num_tokens": [1],
-        "num_blocks": [32],
+        "num_blocks": [4],
+        "block_size": [8],
     },
     "test_gather_cache_mla": {
-        "num_blocks": [32],
-        "max_seq_len": [64],
+        "num_blocks": [4],
+        "block_size": [8],
+        "max_seq_len": [4],
     },
 }
 

tests/test_deepseek_scaling_rope.py

@@ -10,6 +10,11 @@
 MINI_PYTEST_PARAMS = {
     "default": {
         "batch": [1],
+        "seed": [123],
+        "q_num_head,k_num_head": [(16, 1)],
+        "rotary_dim": [64],
+        "q_head_pad,k_head_pad": [(0, 0)],
+        "is_neox": [True],
     },
 }
 

tests/test_fp8_quant.py

@@ -132,8 +132,8 @@ def seed_everything(seed):
 #override pytest parameters when enable mini pytest
 MINI_PYTEST_PARAMS = {
     "default": {
-        "num_tokens": [1, 7, 83],
-        "hidden_size": [1, 2, 3, 4, 16],
+        "num_tokens": [1],
+        "hidden_size": [1],
     },
 }
 

tests/test_grouped_topk.py

@@ -9,7 +9,11 @@
 #override pytest parameters when enable mini pytest
 MINI_PYTEST_PARAMS = {
     "default": {
-        "n_hidden": [128, 256],
+        "n_hidden": [4],
+        "n_token": [1],
+        "routed_scaling_factor": [1.0],
+        "scoring_func": ["softmax"],
+        "renormalize": [True],
     },
 }
 

tests/test_rotary_embedding.py

@@ -34,9 +34,9 @@ def rotary_embedding_opcheck(rot,
 #override pytest parameters when enable mini pytest
 MINI_PYTEST_PARAMS = {
     "default": {
-        "max_position": [11, 256],
+        "max_position": [11],
         "head_size": [32],
-        "seq_len": [11, 128],
+        "seq_len": [11],
     },
 }