kanshan/TensorRT-LLMs
1
0
Fork 0
mirror of https://github.com/NVIDIA/TensorRT-LLM.git synced 2026-01-14 06:27:45 +08:00
Code Issues Actions 1 Packages Projects Releases Wiki Activity

test: fix some test failure and add llama_nemotron models in perf sanity test, add more torch cases (#5693)

Browse Source 
Signed-off-by: ruodil <200874449+ruodil@users.noreply.github.com>
Co-authored-by: Larry <197874197+LarryXFly@users.noreply.github.com>
This commit is contained in:
ruodil 2025-07-07 11:11:41 +08:00 committed by Zhenhuan Chen
parent c8874a7f94
commit 278a1a7df3
4 changed files with 63 additions and 31 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
tests/integration/defs/perf/test_perf.py
View File

@ -64,6 +64,8 @@ MODEL_PATH_DICT = {
"nemotron-nas/Llama-3_3-Nemotron-Super-49B-v1",
"llama_v3.3_nemotron_super_49b_fp8":
"nemotron-nas/Llama-3_3-Nemotron-Super-49B-v1-FP8",
"llama_v3.1_nemotron_ultra_253b":
"nemotron-nas/Llama-3_1-Nemotron-Ultra-253B-v1",
"llama_v3.1_nemotron_ultra_253b_fp8":
"nemotron-nas/Llama-3_1-Nemotron-Ultra-253B-v1-FP8",
"llama_v4_scout_17b_16e_instruct":

1
tests/integration/test_lists/qa/trt_llm_release_perf_l2_test.yml
View File

@ -13,6 +13,7 @@ trt_llm_release_perf_l2_test:
tests:
- perf/test_perf.py::test_perf[deepseek_r1_fp8-bench-pytorch-float8-maxbs:1-input_output_len:1000,2000-reqs:10-con:1-ep:4-tp:8-gpus:8] #min latency test
- perf/test_perf.py::test_perf[deepseek_r1_fp8-bench-pytorch-float8-maxbs:128-maxnt:1127-input_output_len:1000,2000-reqs:5120-con:1024-ep:8-tp:8-gpus:8] #max throughput test
- perf/test_perf.py::test_perf[llama_v3.1_nemotron_nano_8b_fp8-bench-pytorch-float8-maxbs:512-maxnt:20000-input_output_len:20000,2000-reqs:500-con:250]
# FP8 specific tests
- condition:

29
tests/integration/test_lists/qa/trt_llm_release_perf_sanity_test.yml
View File

@ -81,6 +81,9 @@ trt_llm_release_perf_sanity_test:
- perf/test_perf.py::test_perf[llama_v3.1_8b_instruct-bench-bfloat16-input_output_len:512,32-quant:fp8]
- perf/test_perf.py::test_perf[llama_v3.1_8b_instruct_fp8-bench-pytorch-float8-input_output_len:128,128]
- perf/test_perf.py::test_perf[llama_v3.1_8b_instruct_fp8-bench-pytorch-float8-input_output_len:512,32]
- perf/test_perf.py::test_perf[llama_v3.1_nemotron_nano_8b_fp8-bench-pytorch-float8-maxbs:512-maxnt:5000-input_output_len:5000,500-reqs:8-con:1]
- perf/test_perf.py::test_perf[llama_v3.1_nemotron_nano_8b_fp8-bench-pytorch-float8-maxbs:512-input_output_len:500,2000-reqs:8-con:1]
- perf/test_perf.py::test_perf[llama_v3.1_nemotron_nano_8b_fp8-bench-pytorch-float8-maxbs:512-input_output_len:1000,1000-reqs:8-con:1]
# Tests for systems with 2+ GPUs
- condition:
@ -101,6 +104,8 @@ trt_llm_release_perf_sanity_test:
- perf/test_perf.py::test_perf[llama_v3.1_8b_instruct-bench-bfloat16-input_output_len:128,128-quant:int8-gpus:2]
- perf/test_perf.py::test_perf[llama_v3.1_8b-bench-bfloat16-maxbs:256-input_output_len:128,128-gpus:2]
- perf/test_perf.py::test_perf[llama_v3.1_8b_instruct-bench-streaming-bfloat16-input_output_len:128,128-gpus:2]
- perf/test_perf.py::test_perf[llama_v3.1_8b-bench-pytorch-bfloat16-maxbs:256-input_output_len:128,128-gpus:2]
- perf/test_perf.py::test_perf[llama_v3.1_8b_instruct-bench-pytorch-streaming-bfloat16-input_output_len:128,128-gpus:2]
- perf/test_perf.py::test_perf[llama_v3.2_1b-bench-bfloat16-maxbs:1-input_output_len:128,128-reqs:10-gpus:2]
- perf/test_perf.py::test_perf[llama_v3.2_1b-bench-pytorch-bfloat16-input_output_len:128,128-gpus:2]
@ -140,8 +145,8 @@ trt_llm_release_perf_sanity_test:
- '*l40s*'
- '*h20*'
tests:
- perf/test_perf.py::test_perf[llama_v3.1_70b-bench-bfloat16-maxbs:1-input_output_len:128,128-reqs:10-gpus:2]
- perf/test_perf.py::test_perf[mixtral_8x7b_v0.1-bench-float16-input_output_len:128,128-gpus:2]
- perf/test_perf.py::test_perf[llama_v3.1_70b-bench-pytorch-bfloat16-maxbs:1-input_output_len:128,128-reqs:10-gpus:2]
- perf/test_perf.py::test_perf[mixtral_8x7b_v0.1-bench-pytorch-float16-input_output_len:128,128-gpus:2]
# Tests for systems with 4+ GPUs
- condition:
@ -161,6 +166,23 @@ trt_llm_release_perf_sanity_test:
- perf/test_perf.py::test_perf[qwen_14b_chat-cppmanager-ootb_except_mha-float16-input_output_len:128,128-gpus:4]
- perf/test_perf.py::test_perf[starcoder_15.5b-cppmanager-exe-plugin_ifb-float16-maxbs:1-input_output_len:512,200-reqs:10-gpus:4]
# FP8 specific tests
- condition:
terms:
supports_fp8: true
ranges:
system_gpu_count:
gte: 4
wildcards:
gpu:
- '*h100*'
- '*h200*'
- '*l40s*'
- '*h20*'
tests:
- perf/test_perf.py::test_perf[llama_v3.3_nemotron_super_49b-bench-pytorch-bfloat16-maxbs:256-input_output_len:500,2000-reqs:250-con:250-tp:4-gpus:4]
- perf/test_perf.py::test_perf[llama_v3.3_nemotron_super_49b_fp8-bench-pytorch-float8-maxbs:256-input_output_len:500,2000-reqs:250-con:250-tp:4-gpus:4]
# Tests for systems with 8+ GPUs
- condition:
ranges:
@ -200,7 +222,7 @@ trt_llm_release_perf_sanity_test:
tests:
- perf/test_perf.py::test_perf[llama_v3.1_70b-bench-bfloat16-maxbs:1-input_output_len:128,128-quant:fp8-gpus:8]
- perf/test_perf.py::test_perf[llama_v3.1_70b-bench-bfloat16-maxbs:1-input_output_len:512,32-quant:fp8-gpus:8]
- perf/test_perf.py::test_perf[llama_v3.1_70b-bench-pytorch-bfloat16-maxbs:1-input_output_len:512,32-quant:fp8-gpus:8]
- perf/test_perf.py::test_perf[llama_v3.3_70b_instruct_fp8-bench-pytorch-float8-input_output_len:128,128-gpus:8]
- condition:
@ -218,3 +240,4 @@ trt_llm_release_perf_sanity_test:
tests:
- perf/test_perf.py::test_perf[deepseek_v3_lite_fp8-bench-pytorch-float8-input_output_len:128,128]
- perf/test_perf.py::test_perf[deepseek_v3_lite_fp8-bench-streaming-pytorch-float8-input_output_len:128,128]
- perf/test_perf.py::test_perf[qwen3_235b_a22b_fp8-bench-pytorch-float8-input_output_len:128,128-con:256-ep:8-gpus:8]

62
tests/integration/test_lists/qa/trt_llm_release_perf_test.yml
View File

@ -66,7 +66,7 @@ trt_llm_release_perf_test:
# Phi-4-mini-instruct
# cpp
- perf/test_perf.py::test_perf[phi_4_mini_instruct-bench-bfloat16-maxbs:32-input_output_len:5000,500-con:250]
- perf/test_perf.py::test_perf[phi_4_mini_instruct-bench-bfloat16-maxbs:32-maxnt:5000-input_output_len:5000,500-con:250]
- perf/test_perf.py::test_perf[phi_4_mini_instruct-bench-bfloat16-maxbs:32-input_output_len:500,2000-con:250]
- perf/test_perf.py::test_perf[phi_4_mini_instruct-bench-bfloat16-maxbs:32-input_output_len:500,2000-quant:fp8-con:250]
# reduced 'reqs' to fit timeout limit
@ -89,16 +89,16 @@ trt_llm_release_perf_test:
- perf/test_perf.py::test_perf[llama_v3_8b_instruct-cppmanager-exe-plugin_ifb-float16-mp-input_output_len:128,128+512,32] #oom for l40s, l20（cuda_runtime_error）#44, mpi abort on a100 36
- perf/test_perf.py::test_perf[llama_v3_8b_instruct-cppmanager-exe-plugin_ifb-bfloat16-gwp:0.0-input_output_len:128,128+512,32] #oom for l40s, l20, mpi abort on a100 35
- perf/test_perf.py::test_perf[llama_v3_8b_instruct-cppmanager-exe-plugin_ifb-bfloat16-gwp:0.5-input_output_len:128,128+512,32] #oom for l40s, l20
- perf/test_perf.py::test_perf[phi_4_mini_instruct-bench-bfloat16-maxbs:32-input_output_len:5000,500-reqs:10-con:1] # timeout for l20, l40s
- perf/test_perf.py::test_perf[phi_4_mini_instruct-bench-bfloat16-maxbs:32-maxnt:5000-input_output_len:5000,500-reqs:10-con:1] # timeout for l20, l40s
# Llama-3.1-Nemotron-Nano-8B-v1
# cpp backend
- perf/test_perf.py::test_perf[llama_v3.1_nemotron_nano_8b-bench-bfloat16-maxbs:64-input_output_len:500,2000-reqs:8-con:1]
- perf/test_perf.py::test_perf[llama_v3.1_nemotron_nano_8b-bench-bfloat16-maxbs:64-input_output_len:500,2000-quant:fp8-reqs:8-con:1]
- perf/test_perf.py::test_perf[llama_v3.1_nemotron_nano_8b-bench-bfloat16-maxbs:64-input_output_len:1000,1000-reqs:8-con:1]
- perf/test_perf.py::test_perf[llama_v3.1_nemotron_nano_8b-bench-bfloat16-maxbs:64-input_output_len:20000,2000-reqs:8-con:1]
- perf/test_perf.py::test_perf[llama_v3.1_nemotron_nano_8b-bench-bfloat16-maxbs:64-input_output_len:20000,2000-quant:fp8-reqs:8-con:1]
- perf/test_perf.py::test_perf[llama_v3.1_nemotron_nano_8b-bench-bfloat16-maxbs:64-input_output_len:5000,500-con:250]
- perf/test_perf.py::test_perf[llama_v3.1_nemotron_nano_8b-bench-bfloat16-maxbs:64-maxnt:20000-input_output_len:20000,2000-reqs:8-con:1]
- perf/test_perf.py::test_perf[llama_v3.1_nemotron_nano_8b-bench-bfloat16-maxbs:64-maxnt:20000-input_output_len:20000,2000-quant:fp8-reqs:8-con:1]
- perf/test_perf.py::test_perf[llama_v3.1_nemotron_nano_8b-bench-bfloat16-maxbs:64-maxnt:5000-input_output_len:5000,500-con:250]
- perf/test_perf.py::test_perf[llama_v3.1_nemotron_nano_8b-bench-bfloat16-maxbs:64-input_output_len:500,2000-con:250]
- perf/test_perf.py::test_perf[llama_v3.1_nemotron_nano_8b-bench-bfloat16-maxbs:64-input_output_len:1000,1000-con:250]
# pyt backend
@ -115,19 +115,18 @@ trt_llm_release_perf_test:
- perf/test_perf.py::test_perf[llama_v3.1_nemotron_nano_8b_fp8-bench-pytorch-float8-maxbs:512-maxnt:5000-input_output_len:5000,500-reqs:500-con:250]
- perf/test_perf.py::test_perf[llama_v3.1_nemotron_nano_8b_fp8-bench-pytorch-float8-maxbs:512-input_output_len:500,2000-reqs:500-con:250]
- perf/test_perf.py::test_perf[llama_v3.1_nemotron_nano_8b_fp8-bench-pytorch-float8-maxbs:512-input_output_len:1000,1000-reqs:500-con:250]
- perf/test_perf.py::test_perf[llama_v3.1_nemotron_nano_8b_fp8-bench-pytorch-float8-maxbs:512-maxnt:20000-input_output_len:20000,2000-reqs:500-con:250]
#long time llama_nemotron cases
- perf/test_perf.py::test_perf[llama_v3.1_nemotron_nano_8b-bench-bfloat16-maxbs:64-input_output_len:5000,500-reqs:8-con:1] # timeout for l20, l40s, a100
- perf/test_perf.py::test_perf[llama_v3.1_nemotron_nano_8b-bench-bfloat16-maxbs:64-input_output_len:5000,500-quant:fp8-reqs:8-con:1] #timeout for l20, l40s, failed for a100
- perf/test_perf.py::test_perf[llama_v3.1_nemotron_nano_8b-bench-bfloat16-maxbs:64-maxnt:5000-input_output_len:5000,500-reqs:8-con:1] # timeout for l20, l40s, a100
- perf/test_perf.py::test_perf[llama_v3.1_nemotron_nano_8b-bench-bfloat16-maxbs:64-maxnt:5000-input_output_len:5000,500-quant:fp8-reqs:8-con:1] #timeout for l20, l40s, failed for a100
- perf/test_perf.py::test_perf[llama_v3.1_nemotron_nano_8b-bench-bfloat16-maxbs:64-input_output_len:1000,1000-quant:fp8-reqs:8-con:1] # timeout for l20, l40s, failed on a100
- perf/test_perf.py::test_perf[llama_v3.1_nemotron_nano_8b-bench-bfloat16-maxbs:64-input_output_len:5000,500-quant:fp8-con:250] # failed for a100
- perf/test_perf.py::test_perf[llama_v3.1_nemotron_nano_8b-bench-bfloat16-maxbs:64-maxnt:5000-input_output_len:5000,500-quant:fp8-con:250] # failed for a100
- perf/test_perf.py::test_perf[llama_v3.1_nemotron_nano_8b-bench-bfloat16-maxbs:64-input_output_len:500,2000-quant:fp8-con:250] # failed on A100
- perf/test_perf.py::test_perf[llama_v3.1_nemotron_nano_8b-bench-bfloat16-maxbs:64-input_output_len:1000,1000-quant:fp8-con:250] # failed on A100 15
- perf/test_perf.py::test_perf[llama_v3.1_nemotron_nano_8b-bench-bfloat16-maxbs:64-input_output_len:20000,2000-con:250] # timeout for l20, l40s, a100
- perf/test_perf.py::test_perf[llama_v3.1_nemotron_nano_8b-bench-bfloat16-maxbs:64-input_output_len:20000,2000-quant:fp8-con:250] # timeout for l20, l40s, failed on A100
- perf/test_perf.py::test_perf[llama_v3.1_nemotron_nano_8b-bench-pytorch-bfloat16-maxbs:512-input_output_len:5000,500-reqs:500-con:250] # failed for l20, need to extend context token to 5000 for l40s and a100， timeout for h20
- perf/test_perf.py::test_perf[llama_v3.1_nemotron_nano_8b-bench-bfloat16-maxbs:64-maxnt:20000-input_output_len:20000,2000-con:250] # timeout for l20, l40s, a100
- perf/test_perf.py::test_perf[llama_v3.1_nemotron_nano_8b-bench-bfloat16-maxbs:64-maxnt:20000-input_output_len:20000,2000-quant:fp8-con:250] # timeout for l20, l40s, failed on A100
- perf/test_perf.py::test_perf[llama_v3.1_nemotron_nano_8b-bench-pytorch-bfloat16-maxbs:512-maxnt:5000-input_output_len:5000,500-reqs:500-con:250] # failed for l20, need to extend context token to 5000 for l40s and a100， timeout for h20
- perf/test_perf.py::test_perf[llama_v3.1_nemotron_nano_8b-bench-pytorch-bfloat16-maxbs:512-input_output_len:1000,1000-reqs:500-con:250]
- perf/test_perf.py::test_perf[llama_v3.1_nemotron_nano_8b-bench-pytorch-bfloat16-maxbs:512-input_output_len:20000,2000-reqs:500-con:250] #need to extend context token to 20000 for l40s, timeout for h20, a100
- perf/test_perf.py::test_perf[llama_v3.1_nemotron_nano_8b-bench-pytorch-bfloat16-maxbs:512-maxnt:20000-input_output_len:20000,2000-reqs:500-con:250] #need to extend context token to 20000 for l40s, timeout for h20, a100
# deepseek_v3_lite_fp8
- perf/test_perf.py::test_perf[deepseek_v3_lite_fp8-bench-pytorch-float8-input_output_len:128,128]
- perf/test_perf.py::test_perf[deepseek_v3_lite_fp8-bench-pytorch-float8-input_output_len:2000,500]
@ -170,8 +169,8 @@ trt_llm_release_perf_test:
- '*h20*'
tests:
- perf/test_perf.py::test_perf[llama_v3.1_8b_instruct-bench-bfloat16-maxbs:256-input_output_len:1000,1000-quant:fp8] # mabs 256 for L20, L40S
- perf/test_perf.py::test_perf[phi_4_mini_instruct-bench-bfloat16-maxbs:32-input_output_len:5000,500-quant:fp8-reqs:10-con:1]
- perf/test_perf.py::test_perf[phi_4_mini_instruct-bench-bfloat16-maxbs:32-input_output_len:5000,500-quant:fp8-reqs:10-con:250]
- perf/test_perf.py::test_perf[phi_4_mini_instruct-bench-bfloat16-maxbs:32-maxnt:5000-input_output_len:5000,500-quant:fp8-reqs:10-con:1]
- perf/test_perf.py::test_perf[phi_4_mini_instruct-bench-bfloat16-maxbs:32-maxnt:5000-input_output_len:5000,500-quant:fp8-reqs:10-con:250]
- perf/test_perf.py::test_perf[phi_4_mini_instruct-bench-bfloat16-maxbs:32-input_output_len:500,2000-quant:fp8-reqs:10-con:250]
# 2 gpus test
@ -195,8 +194,9 @@ trt_llm_release_perf_test:
- perf/test_perf.py::test_perf[llama_v3.1_8b-bench-streaming-bfloat16-input_output_len:128,128-gpus:2]
- perf/test_perf.py::test_perf[llama_v3.1_8b-bench-bfloat16-maxbs:256-input_output_len:128,128-gpus:2]
- perf/test_perf.py::test_perf[mixtral_8x7b_v0.1-bench-float16-input_output_len:128,128-loras:8-gpus:2]
- perf/test_perf.py::test_perf[llama_v3.2_1b-bench-bfloat16-input_output_len:5000,500-reqs:10-con:1-gpus:2]
- perf/test_perf.py::test_perf[llama_v3.2_1b-bench-bfloat16-input_output_len:5000,500-reqs:10-con:250-gpus:2]
- perf/test_perf.py::test_perf[mixtral_8x7b_v0.1-bench-pytorch-float16-input_output_len:128,128-loras:8-gpus:2]
- perf/test_perf.py::test_perf[llama_v3.2_1b-bench-bfloat16-maxnt:5000-input_output_len:5000,500-reqs:10-con:1-gpus:2]
- perf/test_perf.py::test_perf[llama_v3.2_1b-bench-bfloat16-maxnt:5000-input_output_len:5000,500-reqs:10-con:250-gpus:2]
- perf/test_perf.py::test_perf[llama_v3.2_1b-bench-bfloat16-input_output_len:128,128-gpus:2]
- condition:
@ -218,7 +218,7 @@ trt_llm_release_perf_test:
- perf/test_perf.py::test_perf[mixtral_8x7b-cppmanager-exe-plugin_ifb-float16-mp-input_output_len:128,128+512,32-gpus:2]
- perf/test_perf.py::test_perf[mixtral_8x7b_v0.1-bench-float16-input_output_len:128,128-gpus:2]
- perf/test_perf.py::test_perf[mixtral_8x7b_v0.1-bench-pytorch-float16-input_output_len:128,128-gpus:2]
- perf/test_perf.py::test_perf[mixtral_8x7b_v0.1-bench-streaming-float16-input_output_len:128,128-gpus:2]
- perf/test_perf.py::test_perf[mixtral_8x7b_v0.1-bench-pytorch-streaming-float16-input_output_len:128,128-gpus:2]
# FP8 specific tests
- condition:
@ -236,15 +236,16 @@ trt_llm_release_perf_test:
- '*h20*'
tests:
- perf/test_perf.py::test_perf[llama_v3.2_1b-bench-bfloat16-input_output_len:128,128-quant:fp8-gpus:2]
- perf/test_perf.py::test_perf[llama_v3.2_1b-bench-pytorch-bfloat16-input_output_len:128,128-quant:fp8-gpus:2]
- perf/test_perf.py::test_perf[llama_v3.2_1b-bench-bfloat16-input_output_len:512,32-quant:fp8-gpus:2]
- perf/test_perf.py::test_perf[llama_v3.2_1b-bench-bfloat16-input_output_len:512,200-quant:fp8-gpus:2]
- perf/test_perf.py::test_perf[llama_v3.2_1b-bench-pytorch-bfloat16-input_output_len:512,200-quant:fp8-gpus:2]
- perf/test_perf.py::test_perf[llama_v3.2_1b-bench-pytorch-bfloat16-input_output_len:512,200-gpus:2]
- perf/test_perf.py::test_perf[mixtral_8x7b_v0.1-bench-float16-input_output_len:128,128-quant:fp8-gpus:2]
- perf/test_perf.py::test_perf[mixtral_8x7b_v0.1-bench-float16-input_output_len:512,32-quant:fp8-gpus:2]
- perf/test_perf.py::test_perf[mistral_7b_v0.1-bench-float16-input_output_len:1000,1000-quant:fp8-tp:2]
- perf/test_perf.py::test_perf[mistral_7b_v0.1-bench-pytorch-float16-input_output_len:1000,1000-quant:fp8-tp:2]
- perf/test_perf.py::test_perf[mistral_7b_v0.1-bench-pytorch-float16-input_output_len:1000,1000-tp:2]
- perf/test_perf.py::test_perf[mistral_7b_v0.1-bench-float16-input_output_len:500,2000-quant:fp8-tp:2]
- perf/test_perf.py::test_perf[mistral_7b_v0.1-bench-pytorch-float16-input_output_len:500,2000-quant:fp8-tp:2]
- perf/test_perf.py::test_perf[mistral_7b_v0.1-bench-pytorch-float16-input_output_len:500,2000-tp:2]
- perf/test_perf.py::test_perf[phi_3_mini_128k_instruct-bench-float16-maxbs:128-input_output_len:1000,1000-quant:fp8-tp:2]
- perf/test_perf.py::test_perf[phi_3_mini_128k_instruct-bench-float16-maxbs:128-input_output_len:500,2000-quant:fp8-tp:2]
@ -264,8 +265,10 @@ trt_llm_release_perf_test:
tests:
- perf/test_perf.py::test_perf[mixtral_8x7b_v0.1-bench-float16-input_output_len:128,128-quant:fp8-gpus:2]
- perf/test_perf.py::test_perf[mixtral_8x7b_v0.1-bench-float16-input_output_len:512,32-quant:fp8-gpus:2]
- perf/test_perf.py::test_perf[llama_v3.2_1b-bench-bfloat16-input_output_len:500,2000-quant:fp8-gpus:2]
- perf/test_perf.py::test_perf[llama_v3.2_1b-bench-bfloat16-input_output_len:500,2000-quant:fp8-reqs:10-con:1-gpus:2]
- perf/test_perf.py::test_perf[mixtral_8x7b_v0.1-bench-pytorch-float16-input_output_len:128,128-quant:fp8-gpus:2]
- perf/test_perf.py::test_perf[mixtral_8x7b_v0.1-bench-pytorch-float16-input_output_len:512,32-quant:fp8-gpus:2]
- perf/test_perf.py::test_perf[llama_v3.2_1b-bench-pytorch-bfloat16-input_output_len:500,2000-quant:fp8-gpus:2]
- perf/test_perf.py::test_perf[llama_v3.2_1b-bench-pytorch-bfloat16-input_output_len:500,2000-quant:fp8-reqs:10-con:1-gpus:2]
- perf/test_perf.py::test_perf[llama_v3.2_1b-bench-bfloat16-input_output_len:500,2000-quant:fp8-con:250-gpus:2]
# 4 gpus test
@ -311,12 +314,12 @@ trt_llm_release_perf_test:
- perf/test_perf.py::test_perf[llama_v3.3_70b_instruct_fp8-bench-float8-input_output_len:128,128-tp:4]
# Llama-Nemotron-Super-49B-v3.3
# cpp
- perf/test_perf.py::test_perf[llama_v3.3_nemotron_super_49b-bench-bfloat16-maxbs:64-input_output_len:5000,500-reqs:4-con:1-gpus:4]
- perf/test_perf.py::test_perf[llama_v3.3_nemotron_super_49b-bench-bfloat16-maxbs:64-input_output_len:5000,500-quant:fp8-reqs:4-con:1-gpus:4]
- perf/test_perf.py::test_perf[llama_v3.3_nemotron_super_49b-bench-bfloat16-maxbs:64-maxnt:5000-input_output_len:5000,500-reqs:4-con:1-gpus:4]
- perf/test_perf.py::test_perf[llama_v3.3_nemotron_super_49b-bench-bfloat16-maxbs:64-maxnt:5000-input_output_len:5000,500-quant:fp8-reqs:4-con:1-gpus:4]
- perf/test_perf.py::test_perf[llama_v3.3_nemotron_super_49b-bench-bfloat16-maxbs:64-input_output_len:500,2000-reqs:4-con:1-gpus:4]
- perf/test_perf.py::test_perf[llama_v3.3_nemotron_super_49b-bench-bfloat16-maxbs:64-input_output_len:500,2000-quant:fp8-reqs:4-con:1-gpus:4]
- perf/test_perf.py::test_perf[llama_v3.3_nemotron_super_49b-bench-bfloat16-maxbs:16-input_output_len:5000,500-con:250-gpus:4]
- perf/test_perf.py::test_perf[llama_v3.3_nemotron_super_49b-bench-bfloat16-maxbs:16-input_output_len:5000,500-quant:fp8-con:250-gpus:4]
- perf/test_perf.py::test_perf[llama_v3.3_nemotron_super_49b-bench-bfloat16-maxbs:16-maxnt:5000-input_output_len:5000,500-con:250-gpus:4]
- perf/test_perf.py::test_perf[llama_v3.3_nemotron_super_49b-bench-bfloat16-maxbs:16-maxnt:5000-input_output_len:5000,500-quant:fp8-con:250-gpus:4]
- perf/test_perf.py::test_perf[llama_v3.3_nemotron_super_49b-bench-bfloat16-maxbs:16-input_output_len:500,2000-con:250-gpus:4]
# pyt
# bfloat16
@ -353,7 +356,7 @@ trt_llm_release_perf_test:
- perf/test_perf.py::test_perf[llama_v3.1_70b_instruct-bench-pytorch-bfloat16-input_output_len:2000,200-reqs:64-gpus:8]
- perf/test_perf.py::test_perf[llama_v3.1_70b_instruct-bench-pytorch-streaming-bfloat16-input_output_len:2000,200-reqs:64-gpus:8]
- perf/test_perf.py::test_perf[llama_v3.3_70b_instruct-bench-bfloat16-input_output_len:128,128-gpus:8]
- perf/test_perf.py::test_perf[llama_v3.3_70b_instruct-bench-bfloat16-maxbs:16-input_output_len:5000,500-reqs:64-con:250-gpus:8]
- perf/test_perf.py::test_perf[llama_v3.3_70b_instruct-bench-bfloat16-maxbs:16-maxnt:5000-input_output_len:5000,500-reqs:64-con:250-gpus:8]
- perf/test_perf.py::test_perf[gpt_20b-bench-float16-maxbs:8-input_output_len:128,128-reqs:80-gpus:8]
- perf/test_perf.py::test_perf[gpt_20b-bench-float16-maxbs:8-input_output_len:512,32-reqs:80-gpus:8]
@ -373,6 +376,9 @@ trt_llm_release_perf_test:
- perf/test_perf.py::test_perf[mixtral_8x7b_v0.1_instruct-bench-float16-input_output_len:128,128-reqs:64-gpus:8] # timeout for a100
- perf/test_perf.py::test_perf[mixtral_8x7b_v0.1_instruct-bench-float16-input_output_len:128,128-reqs:10-con:50-gpus:8] # timeout for a100
- perf/test_perf.py::test_perf[mixtral_8x7b_v0.1_instruct-bench-float16-input_output_len:128,128-reqs:10-con:1-gpus:8] # timeout for a100
- perf/test_perf.py::test_perf[mixtral_8x7b_v0.1_instruct-bench-pytorch-float16-input_output_len:128,128-reqs:64-gpus:8] # timeout for a100
- perf/test_perf.py::test_perf[mixtral_8x7b_v0.1_instruct-bench-pytorch-float16-input_output_len:128,128-reqs:10-con:50-gpus:8] # timeout for a100
- perf/test_perf.py::test_perf[mixtral_8x7b_v0.1_instruct-bench-pytorch-float16-input_output_len:128,128-reqs:10-con:1-gpus:8] # timeout for a100
# Llama-3_1-Nemotron-Ultra-253B-v1
# all cpp backend, bf16->fp8 post-quantized
- perf/test_perf.py::test_perf[llama_v3.1_nemotron_ultra_253b-bench-bfloat16-maxbs:64-input_output_len:5000,500-quant:fp8-reqs:8-con:1-tp:8-gpus:8]