[CI] Add Buildkite (#2355)

tests/kernels/test_attention.py

@@ -13,7 +13,7 @@ FLOAT32_BYTES = torch.finfo(torch.float).bits // 8
 # This will change depending on the compute capability.
 # - 512 as a buffer
 MAX_SEQ_LEN = get_max_shared_memory_bytes() // FLOAT32_BYTES - 512
-NUM_BLOCKS = 40000  # Arbitrary values for testing
+NUM_BLOCKS = 12000  # Arbitrary values for testing
 PARTITION_SIZE = 512
 
 DTYPES = [torch.half, torch.bfloat16, torch.float]

tests/kernels/test_cache.py

@@ -6,12 +6,12 @@ import torch
 from vllm._C import cache_ops
 
 DTYPES = [torch.half, torch.bfloat16, torch.float]
-NUM_TOKENS = [83]  # Arbitrary values for testing
+NUM_TOKENS = [42]  # Arbitrary values for testing
 NUM_LAYERS = [1]  # Arbitrary values for testing
 NUM_HEADS = [8]  # Arbitrary values for testing
 HEAD_SIZES = [64, 80, 96, 112, 128, 256]
 BLOCK_SIZES = [8, 16, 32]
-NUM_BLOCKS = [1024, 36000]  # Arbitrary values for testing
+NUM_BLOCKS = [1024, 3600]  # Arbitrary values for testing
 NUM_MAPPINGS = [256]  # Arbitrary values for testing
 SEEDS = [0]
 DEVICES = [i for i in range(1 if torch.cuda.device_count() == 1 else 2)]