A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 14, 05:30 UTC
[UT] enhance free kv cache block queue popleft_n (#24220)
Signed-off-by: Andy Xie <andy.xning@gmail.com>
Ning Xie committed
Sep 15, 2025 at 18:04 UTC
bc0f6059a24cdebc125c984d0ca1138755285a81
1 file changed
+4
tests/v1/core/test_kv_cache_utils.py
+4
@@ -280,9 +280,11 @@ def test_free_kv_cache_block_queue_popleft_n():
280
# Pop 0 block
281
# fake_head->b1->b3->b5->b4->b0->b2->fake_tail
282
assert len(queue.popleft_n(0)) == 0
283
+ assert queue.num_free_blocks == 6
284
# Pop 1 block
285
# fake_head->b3->b5->b4->b0->b2->fake_tail
286
result_blocks = queue.popleft_n(1)
287
+ assert queue.num_free_blocks == 5
288
assert len(result_blocks) == 1
289
assert result_blocks[0] is blocks[1]
290
for block in result_blocks:
@@ -292,6 +294,7 @@ def test_free_kv_cache_block_queue_popleft_n():
294
# fake_head->b4->b0->b2->fake_tail
295
result_blocks = queue.popleft_n(2)
296
assert len(result_blocks) == 2
297
+ assert queue.num_free_blocks == 3
298
assert result_blocks[0] is blocks[3]
299
assert result_blocks[1] is blocks[5]
300
for block in result_blocks:
@@ -301,6 +304,7 @@ def test_free_kv_cache_block_queue_popleft_n():
304
# fake_head->fake_tail
305
result_blocks = queue.popleft_n(3)
306
assert len(result_blocks) == 3
307
+ assert queue.num_free_blocks == 0
308
assert result_blocks[0] is blocks[4]
309
assert result_blocks[1] is blocks[0]
310
assert result_blocks[2] is blocks[2]