[Core] remove cupy dependency (#3625)
This commit is contained in:
@@ -1,13 +1,22 @@
|
||||
"""Compare the outputs of HF and distributed vLLM when using greedy sampling.
|
||||
|
||||
Run `pytest tests/distributed/test_basic_distributed_correctness.py --forked`.
|
||||
vLLM will allocate all the available memory, so we need to run the tests one
|
||||
by one. The solution is to pass arguments (model name) by environment
|
||||
variables.
|
||||
Run:
|
||||
```sh
|
||||
TEST_DIST_MODEL=facebook/opt-125m pytest \
|
||||
test_basic_distributed_correctness.py
|
||||
TEST_DIST_MODEL=meta-llama/Llama-2-7b-hf \
|
||||
test_basic_distributed_correctness.py
|
||||
```
|
||||
"""
|
||||
import os
|
||||
|
||||
import pytest
|
||||
import torch
|
||||
|
||||
MODELS = [
|
||||
"facebook/opt-125m",
|
||||
"meta-llama/Llama-2-7b-hf",
|
||||
os.environ["TEST_DIST_MODEL"],
|
||||
]
|
||||
|
||||
|
||||
|
||||
Reference in New Issue
Block a user