biondizzle/vllm
1
0
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases Wiki Activity

[Core][Model] Terratorch backend integration (#23513)

Browse Source 
Signed-off-by: Michele Gazzetti <michele.gazzetti1@ibm.com>
Signed-off-by: Christian Pinto <christian.pinto@ibm.com>
Co-authored-by: Christian Pinto <christian.pinto@ibm.com>
Co-authored-by: Cyrus Leung <tlleungac@connect.ust.hk>
This commit is contained in:
mgazz
2025-09-04 08:22:41 +01:00
committed by GitHub
parent e7fc70016f
commit 51d5e9be7d
23 changed files with 305 additions and 208 deletions
Download Patch File Download Diff File Expand all files Collapse all files

103
tests/plugins_tests/test_io_processor_plugins.py
View File

@@ -7,12 +7,11 @@ import requests
from tests.utils import RemoteOpenAIServer
from vllm.config import VllmConfig
from vllm.entrypoints.llm import LLM
from vllm.entrypoints.openai.protocol import IOProcessorResponse
from vllm.plugins.io_processors import get_io_processor
from vllm.pooling_params import PoolingParams
MODEL_NAME = "christian-pinto/Prithvi-EO-2.0-300M-TL-VLLM"
MODEL_NAME = "mgazz/Prithvi-EO-2.0-300M-TL-Sen1Floods11"
image_url = "https://huggingface.co/christian-pinto/Prithvi-EO-2.0-300M-TL-VLLM/resolve/main/valencia_example_2024-10-26.tiff" # noqa: E501
@@ -23,61 +22,7 @@ def test_loading_missing_plugin():
get_io_processor(vllm_config, "wrong_plugin")
def test_loading_engine_with_wrong_plugin():
with pytest.raises(ValueError):
LLM(
model=MODEL_NAME,
skip_tokenizer_init=True,
trust_remote_code=True,
enforce_eager=True,
# Limit the maximum number of parallel requests
# to avoid the model going OOM in CI.
max_num_seqs=32,
io_processor_plugin="wrong_plugin",
)
@pytest.mark.parametrize("model_name", [MODEL_NAME])
def test_prithvi_mae_plugin_offline(vllm_runner, model_name: str):
img_prompt = dict(
data=image_url,
data_format="url",
image_format="tiff",
out_data_format="b64_json",
)
pooling_params = PoolingParams(task="encode", softmax=False)
with vllm_runner(
model_name,
runner="pooling",
skip_tokenizer_init=True,
trust_remote_code=True,
enforce_eager=True,
# Limit the maximum number of parallel requests
# to avoid the model going OOM in CI.
max_num_seqs=1,
io_processor_plugin="prithvi_to_tiff_valencia",
) as llm_runner:
pooler_output = llm_runner.get_llm().encode(
img_prompt,
pooling_params=pooling_params,
)
output = pooler_output[0].outputs
# verify the output is formatted as expected for this plugin
assert all(
hasattr(output, attr)
for attr in ["type", "format", "data", "request_id"])
# We just check that the output is a valid base64 string.
# Raises an exception and fails the test if the string is corrupted.
base64.b64decode(output.data)
@pytest.fixture(scope="module")
@pytest.fixture(scope="function")
def server():
args = [
"--runner",
@@ -90,7 +35,9 @@ def server():
"--max-num-seqs",
"32",
"--io-processor-plugin",
"prithvi_to_tiff_valencia"
"prithvi_to_tiff_valencia",
"--model-impl",
"terratorch",
]
with RemoteOpenAIServer(MODEL_NAME, args) as remote_server:
@@ -136,3 +83,43 @@ async def test_prithvi_mae_plugin_online(
# We just check that the output is a valid base64 string.
# Raises an exception and fails the test if the string is corrupted.
base64.b64decode(plugin_data["data"])
@pytest.mark.parametrize("model_name", [MODEL_NAME])
def test_prithvi_mae_plugin_offline(vllm_runner, model_name: str):
img_prompt = dict(
data=image_url,
data_format="url",
image_format="tiff",
out_data_format="b64_json",
)
pooling_params = PoolingParams(task="encode", softmax=False)
with vllm_runner(
model_name,
runner="pooling",
skip_tokenizer_init=True,
trust_remote_code=True,
enforce_eager=True,
# Limit the maximum number of parallel requests
# to avoid the model going OOM in CI.
max_num_seqs=1,
model_impl="terratorch",
io_processor_plugin="prithvi_to_tiff_valencia",
) as llm_runner:
pooler_output = llm_runner.get_llm().encode(
img_prompt,
pooling_params=pooling_params,
)
output = pooler_output[0].outputs
# verify the output is formatted as expected for this plugin
assert all(
hasattr(output, attr)
for attr in ["type", "format", "data", "request_id"])
# We just check that the output is a valid base64 string.
# Raises an exception and fails the test if the string is corrupted.
base64.b64decode(output.data)