Unverified Commit da7d3e9e authored by knarangN's avatar knarangN Committed by GitHub
Browse files

test: add TensorRT-LLM multimodal EPD test for nightly CI (#6193)


Signed-off-by: default avatarKavita Narang <knarang@nvidia.com>
parent f91b42b9
...@@ -199,21 +199,37 @@ trtllm_configs = { ...@@ -199,21 +199,37 @@ trtllm_configs = {
delayed_start=60, delayed_start=60,
request_payloads=[multimodal_payload_default()], request_payloads=[multimodal_payload_default()],
), ),
"epd_multimodal_image_and_embeddings": TRTLLMConfig( # TensorRT-LLM EPD (Encode-Prefill-Decode) multimodal test for nightly CI
name="epd_multimodal_image_and_embeddings", # Uses llava model with 2 GPUs (encode shares GPU with prefill)
#
# TODO: Add Llama-4-Scout multimodal tests (agg_multimodal_llama, disagg_multimodal_llama)
# once CI supports gpu_8 runners and launch scripts are available
"epd_multimodal": TRTLLMConfig(
name="epd_multimodal",
directory=trtllm_dir, directory=trtllm_dir,
script_name="epd_multimodal_image_and_embeddings.sh", script_name="epd_multimodal_image.sh",
marks=[ marks=[
pytest.mark.gpu_4, pytest.mark.gpu_2,
pytest.mark.trtllm, pytest.mark.trtllm,
pytest.mark.multimodal, pytest.mark.multimodal,
pytest.mark.nightly, pytest.mark.nightly,
], ],
model="llava-hf/llava-v1.6-mistral-7b-hf", model="llava-hf/llava-v1.6-mistral-7b-hf",
frontend_port=DefaultPort.FRONTEND.value, frontend_port=DefaultPort.FRONTEND.value,
timeout=1200, timeout=900,
delayed_start=120, delayed_start=120,
request_payloads=[multimodal_payload_default()], request_payloads=[
multimodal_payload_default(
text="Describe what you see in this image.",
expected_response=["mountain", "rock", "trees", "road"],
)
],
env={
# Override GPU assignments to fit on 2 GPUs (encode shares with prefill)
"PREFILL_CUDA_VISIBLE_DEVICES": "0",
"DECODE_CUDA_VISIBLE_DEVICES": "1",
"ENCODE_CUDA_VISIBLE_DEVICES": "0",
},
), ),
"completions_only": TRTLLMConfig( "completions_only": TRTLLMConfig(
name="completions_only", name="completions_only",
......
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment