kanshan/TensorRT-LLMs
1
0
Fork 0
mirror of https://github.com/NVIDIA/TensorRT-LLM.git synced 2026-01-13 22:18:36 +08:00
Code Issues Actions 1 Packages Projects Releases Wiki Activity

[TRTLLM-9381][test] add disag-serving kimi k2 thinking tests (#10357)

Browse Source 
Signed-off-by: Xin He (SW-GPU) <200704525+xinhe-nv@users.noreply.github.com>
This commit is contained in:
xinhe-nv 2026-01-05 18:15:52 +08:00 committed by GitHub
parent 4931c5eb3a
commit b1733d56f6
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
3 changed files with 60 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
tests/integration/defs/accuracy/references/gsm8k.yaml
View File

@ -150,6 +150,8 @@ moonshotai/Kimi-K2-Thinking:
- quant_algo: NVFP4
kv_cache_quant_algo: FP8
accuracy: 90.84
- quant_algo: NVFP4
accuracy: 90.84
nvidia/Llama-3_3-Nemotron-Super-49B-v1:
- accuracy: 92.57
- quant_algo: FP8

57
tests/integration/defs/accuracy/test_disaggregated_serving.py
View File

@ -1339,3 +1339,60 @@ class TestQwen3_30B_A3B(LlmapiAccuracyTestHarness):
gen_model=gen_model,
ctx_instances=1,
gen_instances=1)
@pytest.mark.timeout(10800)
@skip_pre_blackwell
class TestKimiK2(LlmapiAccuracyTestHarness):
MODEL_NAME = "moonshotai/Kimi-K2-Instruct"
MODEL_PATH = f"{llm_models_root()}/Kimi-K2-Instruct"
@pytest.mark.skip_less_device(8)
@pytest.mark.skip_less_device_memory(200000)
def test_nvfp4(self):
model_name = "moonshotai/Kimi-K2-Thinking"
model_path = f"{llm_models_root()}/Kimi-K2-Thinking-NVFP4"
ctx_server_config = {
"max_batch_size": 16,
"disable_overlap_scheduler": True,
"cache_transceiver_config": {
"backend": "DEFAULT"
},
"tensor_parallel_size": 4,
"enable_attention_dp": True,
"trust_remote_code": True,
"kv_cache_config": {
"free_gpu_memory_fraction": 0.8,
},
}
gen_server_config = {
"max_batch_size": 16,
"disable_overlap_scheduler": True,
"cache_transceiver_config": {
"backend": "DEFAULT"
},
"tensor_parallel_size": 4,
"enable_attention_dp": True,
"trust_remote_code": True,
"kv_cache_config": {
"free_gpu_memory_fraction": 0.8,
},
}
disaggregated_server_config = {
"hostname": "localhost",
"port": 8000,
"backend": "pytorch",
"context_servers": {
"num_instances": 1,
"urls": ["localhost:8001"]
},
"generation_servers": {
"num_instances": 1,
"urls": ["localhost:8002"]
}
}
with launch_disaggregated_llm(disaggregated_server_config,
ctx_server_config, gen_server_config,
model_path) as llm:
task = GSM8K(model_name)
task.evaluate(llm)

1
tests/integration/test_lists/qa/llm_function_core.txt
View File

@ -637,6 +637,7 @@ accuracy/test_disaggregated_serving.py::TestLlama4ScoutInstruct::test_auto_dtype
accuracy/test_disaggregated_serving.py::TestLlama4ScoutInstruct::test_auto_dtype[True]
accuracy/test_disaggregated_serving.py::TestQwen3_8B::test_auto_dtype[False]
accuracy/test_disaggregated_serving.py::TestQwen3_8B::test_auto_dtype[True]
accuracy/test_disaggregated_serving.py::TestKimiK2::test_nvfp4
accuracy/test_llm_api_pytorch.py::TestQwen3_235B_A22B::test_fp8[throughput_latency]
accuracy/test_llm_api_pytorch.py::TestQwen3_235B_A22B::test_fp8[latency]
accuracy/test_llm_api_pytorch.py::TestQwen3_235B_A22B::test_nvfp4[latency_moe_cutlass]