@tduncan / Nvidia-Optimization-Stack-VLLM Mirror · read-only / commits / fcc2994be6
Fork 0

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 17, 01:45 UTC

[CI] Nits for bad initialization of SeqGroup in testing (#4748)

Robert Shaw committed May 10, 2024 at 16:01 UTC fcc2994be657a897dd0732928754749048520b28
2 files changed +16 -8
tests/core/test_block_manager.py
+9 -4
@@ -142,8 +142,10 @@ def test_append_slot_cow():
142 child = prompt.fork(new_seq_id=2)
143
144 # Allocate space for the sequence group.
145 - seq_group = SequenceGroup("1", [prompt, child], SamplingParams(),
146 - time.time(), time.perf_counter)
145 + seq_group = SequenceGroup(request_id="1",
146 + seqs=[prompt, child],
147 + arrival_time=time.time(),
148 + sampling_params=SamplingParams())
149 block_manager.allocate(seq_group)
150
151 # Fork and append a new token id. We expect a COW to be scheduled.
@@ -303,8 +305,11 @@ def test_sliding_window_multi_seq():
305 assert block_manager.get_num_free_gpu_blocks() == num_gpu_blocks
306
307 parent = Sequence(1, "one two three", [0, 1, 2], block_size)
306 - seq_group = SequenceGroup("1", [parent], SamplingParams(), time.time(),
307 - None)
308 + seq_group = SequenceGroup(request_id="1",
309 + seqs=[parent],
310 + arrival_time=time.time(),
311 + sampling_params=SamplingParams(),
312 + lora_request=None)
313 block_manager.allocate(seq_group)
314
315 # assert the number of blocks allocated is correct
tests/core/utils.py
+7 -4
@@ -22,10 +22,13 @@ def create_dummy_prompt(
22 prompt_tokens = list(range(prompt_length))
23 prompt_str = " ".join([str(t) for t in prompt_tokens])
24 prompt = Sequence(int(request_id), prompt_str, prompt_tokens, block_size)
25 - seq_group = SequenceGroup(
26 - request_id, [prompt],
27 - SamplingParams(use_beam_search=use_beam_search, best_of=best_of),
28 - time.time(), lora_request)
25 + seq_group = SequenceGroup(request_id=request_id,
26 + seqs=[prompt],
27 + arrival_time=time.time(),
28 + sampling_params=SamplingParams(
29 + use_beam_search=use_beam_search,
30 + best_of=best_of),
31 + lora_request=lora_request)
32
33 return prompt, seq_group
34