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
[Misc] Remove cruft file in repo (#25678)
Signed-off-by: NickLucche <nlucches@redhat.com>
Nicolò Lucchesi committed
Sep 25, 2025 at 17:05 UTC
0754ac4c492776f33d6092391ad2bf4a9dc32925
1 file changed
-61
tests/test_test.py
deleted
-61
@@ -1,61 +0,0 @@
1
-# SPDX-License-Identifier: Apache-2.0
2
-# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
3
-
4
-import pytest
5
-
6
-from vllm import LLM, envs
7
-from vllm.sampling_params import SamplingParams
8
-
9
-if not envs.VLLM_USE_V1:
10
- pytest.skip(
11
- "Skipping V1 tests. Rerun with `VLLM_USE_V1=1` to test.",
12
- allow_module_level=True,
13
- )
14
-
15
-
16
-@pytest.mark.parametrize("model_name", ["Qwen/Qwen2.5-1.5B-Instruct"])
17
-# TODO TPU will appear busy if we fan-out test params here
18
-@pytest.mark.parametrize("n_prompts", [1])
19
-def test_logprobs(model_name: str, n_prompts: int):
20
- """
21
- Request top logprobs with different sampling settings and check
22
- that results contains the requested number, ordered ascendingly.
23
- """
24
-
25
- def check_num_logprobs(logprobs, expected_num: int):
26
- for step in logprobs:
27
- prev_logp = 1.0
28
- # order by rank
29
- sorted_step = dict(
30
- sorted(step.items(), key=lambda item: item[1].rank))
31
-
32
- if len(step) != expected_num:
33
- print("watch out", sorted_step)
34
-
35
- # check results are ordered by prob value
36
- # assert len(step) == expected_num
37
- for rankno, (tid, logp) in enumerate(sorted_step.items()):
38
- assert logp.logprob <= prev_logp
39
- prev_logp = logp.logprob
40
- assert logp.rank == rankno + 1
41
-
42
- llm = LLM(model_name,
43
- enforce_eager=False,
44
- max_num_seqs=1,
45
- max_model_len=128,
46
- max_num_batched_tokens=128)
47
- prompts = [
48
- "Write a short story about a robot that dreams for the first time."
49
- ] * n_prompts
50
- greedy_sampling_params = SamplingParams(temperature=0.0, max_tokens=64,\
51
- logprobs=4)
52
- regular_sampling_params = SamplingParams(temperature=0.4, max_tokens=64,\
53
- logprobs=4)
54
- topkp_sampling_params = SamplingParams(temperature=0.4, max_tokens=64,\
55
- logprobs=4, top_k=12, top_p=0.5)
56
-
57
- for sp in [greedy_sampling_params, regular_sampling_params, \
58
- topkp_sampling_params]:
59
- output = llm.generate(prompts, sp)
60
- for o in output:
61
- check_num_logprobs(o.outputs[0].logprobs, 4)