vbatts/llama.cpp
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

tool-call: update scripts/fetch_server_test_models.py

Browse source
This commit is contained in:
ochafik 2024-10-28 23:57:23 +00:00
parent b825440c81
commit aefac1e5cb
2 changed files with 22 additions and 4 deletions
Download patch file Download diff file Expand all files Collapse all files

7
examples/server/tests/README.md
View file

@ -62,3 +62,10 @@ After changing logic in `steps.py`, ensure that `@bug` and `@wrong_usage` scenar
```shell ```shell
./tests.sh --no-skipped --tags bug,wrong_usage || echo "should failed but compile" ./tests.sh --no-skipped --tags bug,wrong_usage || echo "should failed but compile"
``` ```
Some tests (especially `@slow` ones) require model downloads. Since this can time out the tests, you can pre-download them in the cache ahead of time with:
```shell
pip install -r examples/server/tests/requirements.txt
python scripts/fetch_server_test_models.py
```

19
scripts/fetch_server_test_models.py
View file

@ -9,12 +9,13 @@
python scripts/fetch_server_test_models.py python scripts/fetch_server_test_models.py
( cd examples/server/tests && ./tests.sh --tags=slow ) ( cd examples/server/tests && ./tests.sh --tags=slow )
''' '''
import os
from behave.parser import Parser from behave.parser import Parser
import glob import glob
import re import os
from pydantic import BaseModel from pydantic import BaseModel
import re
import subprocess import subprocess
import sys
class HuggingFaceModel(BaseModel): class HuggingFaceModel(BaseModel):
@ -60,8 +61,18 @@ cli_path = os.environ.get(
os.path.dirname(__file__), os.path.dirname(__file__),
'../build/bin/Release/llama-cli.exe' if os.name == 'nt' else '../build/bin/llama-cli')) '../build/bin/Release/llama-cli.exe' if os.name == 'nt' else '../build/bin/llama-cli'))
for m in models: for m in sorted(list(models), key=lambda m: m.hf_repo):
if '<' in m.hf_repo or '<' in m.hf_file: if '<' in m.hf_repo or '<' in m.hf_file:
continue continue
if '-of-' in m.hf_file:
print(f'# Skipping model at {m.hf_repo} / {m.hf_file} because it is a split file', file=sys.stderr)
continue
print(f'# Ensuring model at {m.hf_repo} / {m.hf_file} is fetched') print(f'# Ensuring model at {m.hf_repo} / {m.hf_file} is fetched')
subprocess.check_call([cli_path, '-hfr', m.hf_repo, '-hff', m.hf_file, '-fa', '-n', '1', '-p', 'Hey', '--no-warmup']) cmd = [cli_path, '-hfr', m.hf_repo, '-hff', m.hf_file, '-n', '1', '-p', 'Hey', '--no-warmup', '--log-disable']
if m.hf_file != 'tinyllamas/stories260K.gguf':
cmd.append('-fa')
try:
subprocess.check_call(cmd)
except subprocess.CalledProcessError:
print(f'# Failed to fetch model at {m.hf_repo} / {m.hf_file} with command:\n {" ".join(cmd)}', file=sys.stderr)
exit(1)