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 15, 03:30 UTC
[CI/Build] Enable phi2 lora test (#20540)
Signed-off-by: Jee Jee Li <pandaleefree@gmail.com>
Jee Jee Li committed
Jul 7, 2025 at 13:10 UTC
2e610deb72dfc1e34b904d9b6b02c85eefa451d2
1 file changed
-5
tests/lora/test_phi.py
-5
@@ -1,8 +1,6 @@
1
# SPDX-License-Identifier: Apache-2.0
2
# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
3
4
-import pytest
5
-
4
import vllm
5
from vllm.lora.request import LoRARequest
6
@@ -49,9 +47,6 @@ def do_sample(llm: vllm.LLM, lora_path: str, lora_id: int) -> list[str]:
47
return generated_texts
48
49
52
-# Skipping for V1 for now as we are hitting,
53
-# "Head size 80 is not supported by FlashAttention." error.
54
-@pytest.mark.skip(reason="Head size 80 is not supported by FlashAttention")
50
def test_phi2_lora(phi2_lora_files):
51
# We enable enforce_eager=True here to reduce VRAM usage for lora-test CI,
52
# Otherwise, the lora-test will fail due to CUDA OOM.