kanshan/TensorRT-LLMs
1
0
Fork 0
mirror of https://github.com/NVIDIA/TensorRT-LLM.git synced 2026-02-05 02:31:33 +08:00
Code Issues Actions 1 Packages Projects Releases Wiki Activity

[None][test] modify ctx config in 128k8k disagg cases (#10779)

Browse Source 
Signed-off-by: Ruodi Lu <ruodil@users.noreply.github.com>
Co-authored-by: Ruodi Lu <ruodil@users.noreply.github.com>
Signed-off-by: Wangshanshan <30051912+dominicshanshan@users.noreply.github.com>
This commit is contained in:
ruodil 2026-01-19 14:10:19 +08:00 committed by Yanchao Lu
parent af49fbdf65
commit 4df0ca8bd1
35 changed files with 60 additions and 59 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
tests/integration/defs/perf/disagg/test_configs/disagg/perf/deepseek-r1-fp4_128k8k_ctx1_pp4_gen13_tep4_bs1_eplb0_mtp0-Default.yaml
View File

@ -83,7 +83,7 @@ worker_config:
max_seq_len: 131104
tensor_parallel_size: 1
moe_expert_parallel_size: 1
enable_attention_dp: true
enable_attention_dp: false
pipeline_parallel_size: 4
print_iter_log: true
cuda_graph_config: null

4
tests/integration/defs/perf/disagg/test_configs/disagg/perf/deepseek-r1-fp4_128k8k_ctx1_pp4_gen5_tep4_bs4_eplb0_mtp0-Default.yaml
View File

@ -78,12 +78,12 @@ worker_config:
num_postprocess_workers: 4
allreduce_strategy: MNNVL
ctx:
max_batch_size: 4
max_batch_size: 1
max_num_tokens: 131104
max_seq_len: 131104
tensor_parallel_size: 1
moe_expert_parallel_size: 1
enable_attention_dp: true
enable_attention_dp: false
pipeline_parallel_size: 4
print_iter_log: true
cuda_graph_config: null

2
tests/integration/defs/perf/disagg/test_configs/disagg/perf/deepseek-r1-fp4_128k8k_ctx1_pp4_gen6_tep8_bs1_eplb0_mtp3-Default.yaml
View File

@ -87,7 +87,7 @@ worker_config:
max_seq_len: 131104
tensor_parallel_size: 1
moe_expert_parallel_size: 1
enable_attention_dp: true
enable_attention_dp: false
pipeline_parallel_size: 4
print_iter_log: true
cuda_graph_config: null

2
tests/integration/defs/perf/disagg/test_configs/disagg/perf/deepseek-r1-fp4_128k8k_ctx1_pp4_gen7_tep8_bs1_eplb0_mtp0-Default.yaml
View File

@ -83,7 +83,7 @@ worker_config:
max_seq_len: 131104
tensor_parallel_size: 1
moe_expert_parallel_size: 1
enable_attention_dp: true
enable_attention_dp: false
pipeline_parallel_size: 4
print_iter_log: true
cuda_graph_config: null

4
tests/integration/defs/perf/disagg/test_configs/disagg/perf/deepseek-r1-fp4_128k8k_ctx1_pp4_gen8_tep4_bs2_eplb0_mtp0-Default.yaml
View File

@ -78,12 +78,12 @@ worker_config:
num_postprocess_workers: 4
allreduce_strategy: MNNVL
ctx:
max_batch_size: 2
max_batch_size: 1
max_num_tokens: 131104
max_seq_len: 131104
tensor_parallel_size: 1
moe_expert_parallel_size: 1
enable_attention_dp: true
enable_attention_dp: false
pipeline_parallel_size: 4
print_iter_log: true
cuda_graph_config: null

2
tests/integration/defs/perf/disagg/test_configs/disagg/perf/deepseek-r1-fp4_128k8k_ctx1_pp4_gen8_tep8_bs1_eplb0_mtp0-Default.yaml
View File

@ -83,7 +83,7 @@ worker_config:
max_seq_len: 131104
tensor_parallel_size: 1
moe_expert_parallel_size: 1
enable_attention_dp: true
enable_attention_dp: false
pipeline_parallel_size: 4
print_iter_log: true
cuda_graph_config: null

4
tests/integration/defs/perf/disagg/test_configs/disagg/perf/deepseek-r1-fp4_128k8k_ctx1_pp8_gen11_tep4_bs2_eplb0_mtp0-Default.yaml
View File

@ -78,12 +78,12 @@ worker_config:
num_postprocess_workers: 4
allreduce_strategy: MNNVL
ctx:
max_batch_size: 2
max_batch_size: 1
max_num_tokens: 131104
max_seq_len: 131104
tensor_parallel_size: 1
moe_expert_parallel_size: 1
enable_attention_dp: true
enable_attention_dp: false
pipeline_parallel_size: 8
print_iter_log: true
cuda_graph_config: null

2
tests/integration/defs/perf/disagg/test_configs/disagg/perf/deepseek-r1-fp4_128k8k_ctx1_pp8_gen14_tep4_bs1_eplb0_mtp0-Default.yaml
View File

@ -83,7 +83,7 @@ worker_config:
max_seq_len: 131104
tensor_parallel_size: 1
moe_expert_parallel_size: 1
enable_attention_dp: true
enable_attention_dp: false
pipeline_parallel_size: 8
print_iter_log: true
cuda_graph_config: null

2
tests/integration/defs/perf/disagg/test_configs/disagg/perf/deepseek-r1-fp4_128k8k_ctx1_pp8_gen1_dep16_bs1_eplb0_mtp3-Default.yaml
View File

@ -86,7 +86,7 @@ worker_config:
max_seq_len: 131104
tensor_parallel_size: 1
moe_expert_parallel_size: 1
enable_attention_dp: true
enable_attention_dp: false
pipeline_parallel_size: 8
print_iter_log: true
cuda_graph_config: null

4
tests/integration/defs/perf/disagg/test_configs/disagg/perf/deepseek-r1-fp4_128k8k_ctx1_pp8_gen1_dep8_bs4_eplb0_mtp2-Default.yaml
View File

@ -81,12 +81,12 @@ worker_config:
decoding_type: MTP
num_nextn_predict_layers: 2
ctx:
max_batch_size: 4
max_batch_size: 1
max_num_tokens: 131104
max_seq_len: 131104
tensor_parallel_size: 1
moe_expert_parallel_size: 1
enable_attention_dp: true
enable_attention_dp: false
pipeline_parallel_size: 8
print_iter_log: true
cuda_graph_config: null

2
tests/integration/defs/perf/disagg/test_configs/disagg/perf/deepseek-r1-fp4_128k8k_ctx1_pp8_gen1_tep8_bs1_eplb0_mtp0-Default.yaml
View File

@ -83,7 +83,7 @@ worker_config:
max_seq_len: 131104
tensor_parallel_size: 1
moe_expert_parallel_size: 1
enable_attention_dp: true
enable_attention_dp: false
pipeline_parallel_size: 8
print_iter_log: true
cuda_graph_config: null

2
tests/integration/defs/perf/disagg/test_configs/disagg/perf/deepseek-r1-fp4_128k8k_ctx1_pp8_gen1_tep8_bs1_eplb0_mtp3-Default.yaml
View File

@ -87,7 +87,7 @@ worker_config:
max_seq_len: 131104
tensor_parallel_size: 1
moe_expert_parallel_size: 1
enable_attention_dp: true
enable_attention_dp: false
pipeline_parallel_size: 8
print_iter_log: true
cuda_graph_config: null

4
tests/integration/defs/perf/disagg/test_configs/disagg/perf/deepseek-r1-fp4_128k8k_ctx1_pp8_gen1_tep8_bs2_eplb0_mtp3-Default.yaml
View File

@ -82,12 +82,12 @@ worker_config:
decoding_type: MTP
num_nextn_predict_layers: 3
ctx:
max_batch_size: 2
max_batch_size: 1
max_num_tokens: 131104
max_seq_len: 131104
tensor_parallel_size: 1
moe_expert_parallel_size: 1
enable_attention_dp: true
enable_attention_dp: false
pipeline_parallel_size: 8
print_iter_log: true
cuda_graph_config: null

4
tests/integration/defs/perf/disagg/test_configs/disagg/perf/deepseek-r1-fp4_128k8k_ctx1_pp8_gen5_tep8_bs2_eplb0_mtp3-Default.yaml
View File

@ -81,12 +81,12 @@ worker_config:
decoding_type: MTP
num_nextn_predict_layers: 3
ctx:
max_batch_size: 2
max_batch_size: 1
max_num_tokens: 131104
max_seq_len: 131104
tensor_parallel_size: 1
moe_expert_parallel_size: 1
enable_attention_dp: true
enable_attention_dp: false
pipeline_parallel_size: 8
print_iter_log: true
cuda_graph_config: null

4
tests/integration/defs/perf/disagg/test_configs/disagg/perf/deepseek-r1-fp4_128k8k_ctx1_pp8_gen7_tep4_bs2_eplb0_mtp2-Default.yaml
View File

@ -81,12 +81,12 @@ worker_config:
decoding_type: MTP
num_nextn_predict_layers: 2
ctx:
max_batch_size: 2
max_batch_size: 1
max_num_tokens: 131104
max_seq_len: 131104
tensor_parallel_size: 1
moe_expert_parallel_size: 1
enable_attention_dp: true
enable_attention_dp: false
pipeline_parallel_size: 8
print_iter_log: true
cuda_graph_config: null

2
tests/integration/defs/perf/disagg/test_configs/disagg/perf/deepseek-r1-fp4_128k8k_ctx1_pp8_gen7_tep8_bs1_eplb0_mtp0-Default.yaml
View File

@ -83,7 +83,7 @@ worker_config:
max_seq_len: 131104
tensor_parallel_size: 1
moe_expert_parallel_size: 1
enable_attention_dp: true
enable_attention_dp: false
pipeline_parallel_size: 8
print_iter_log: true
cuda_graph_config: null

4
tests/integration/defs/perf/disagg/test_configs/disagg/perf/deepseek-r1-fp4_128k8k_ctx1_pp8_gen8_tep4_bs4_eplb0_mtp0-Default.yaml
View File

@ -78,12 +78,12 @@ worker_config:
num_postprocess_workers: 4
allreduce_strategy: MNNVL
ctx:
max_batch_size: 4
max_batch_size: 1
max_num_tokens: 131104
max_seq_len: 131104
tensor_parallel_size: 1
moe_expert_parallel_size: 1
enable_attention_dp: true
enable_attention_dp: false
pipeline_parallel_size: 8
print_iter_log: true
cuda_graph_config: null

4
tests/integration/defs/perf/disagg/test_configs/disagg/perf/deepseek-r1-fp4_128k8k_ctx2_pp4_gen7_tep8_bs2_eplb0_mtp3-Default.yaml
View File

@ -81,12 +81,12 @@ worker_config:
num_nextn_predict_layers: 3
allreduce_strategy: MNNVL
ctx:
max_batch_size: 2
max_batch_size: 1
max_num_tokens: 131104
max_seq_len: 131104
tensor_parallel_size: 1
moe_expert_parallel_size: 1
enable_attention_dp: true
enable_attention_dp: false
pipeline_parallel_size: 4
print_iter_log: true
cuda_graph_config: null

4
tests/integration/defs/perf/disagg/test_configs/disagg/perf/deepseek-r1-fp4_128k8k_ctx2_pp8_gen1_dep32_bs2_eplb0_mtp0-Default.yaml
View File

@ -77,12 +77,12 @@ worker_config:
stream_interval: 20
num_postprocess_workers: 4
ctx:
max_batch_size: 2
max_batch_size: 1
max_num_tokens: 131104
max_seq_len: 131104
tensor_parallel_size: 1
moe_expert_parallel_size: 1
enable_attention_dp: true
enable_attention_dp: false
pipeline_parallel_size: 8
print_iter_log: true
cuda_graph_config: null

4
tests/integration/defs/perf/disagg/test_configs/disagg/perf/deepseek-r1-fp4_128k8k_ctx3_pp4_gen1_dep8_bs16_eplb0_mtp1-Default.yaml
View File

@ -80,12 +80,12 @@ worker_config:
decoding_type: MTP
num_nextn_predict_layers: 1
ctx:
max_batch_size: 16
max_batch_size: 1
max_num_tokens: 131104
max_seq_len: 131104
tensor_parallel_size: 1
moe_expert_parallel_size: 1
enable_attention_dp: true
enable_attention_dp: false
pipeline_parallel_size: 4
print_iter_log: true
cuda_graph_config: null

4
tests/integration/defs/perf/disagg/test_configs/disagg/perf/deepseek-r1-fp4_128k8k_ctx3_pp8_gen1_dep16_bs16_eplb0_mtp0-Default.yaml
View File

@ -77,12 +77,12 @@ worker_config:
stream_interval: 20
num_postprocess_workers: 4
ctx:
max_batch_size: 16
max_batch_size: 1
max_num_tokens: 131104
max_seq_len: 131104
tensor_parallel_size: 1
moe_expert_parallel_size: 1
enable_attention_dp: true
enable_attention_dp: false
pipeline_parallel_size: 8
print_iter_log: true
cuda_graph_config: null

4
tests/integration/defs/perf/disagg/test_configs/disagg/perf/deepseek-r1-fp4_128k8k_ctx3_pp8_gen1_dep16_bs8_eplb0_mtp2-Default.yaml
View File

@ -81,12 +81,12 @@ worker_config:
decoding_type: MTP
num_nextn_predict_layers: 2
ctx:
max_batch_size: 8
max_batch_size: 1
max_num_tokens: 131104
max_seq_len: 131104
tensor_parallel_size: 1
moe_expert_parallel_size: 1
enable_attention_dp: true
enable_attention_dp: false
pipeline_parallel_size: 8
print_iter_log: true
cuda_graph_config: null

4
tests/integration/defs/perf/disagg/test_configs/disagg/perf/deepseek-r1-fp4_128k8k_ctx3_pp8_gen1_dep32_bs2_eplb0_mtp3-Default.yaml
View File

@ -81,12 +81,12 @@ worker_config:
decoding_type: MTP
num_nextn_predict_layers: 3
ctx:
max_batch_size: 2
max_batch_size: 1
max_num_tokens: 131104
max_seq_len: 131104
tensor_parallel_size: 1
moe_expert_parallel_size: 1
enable_attention_dp: true
enable_attention_dp: false
pipeline_parallel_size: 8
print_iter_log: true
cuda_graph_config: null

4
tests/integration/defs/perf/disagg/test_configs/disagg/perf/deepseek-r1-fp4_128k8k_ctx3_pp8_gen1_dep32_bs4_eplb0_mtp0-Default.yaml
View File

@ -77,12 +77,12 @@ worker_config:
stream_interval: 20
num_postprocess_workers: 4
ctx:
max_batch_size: 4
max_batch_size: 1
max_num_tokens: 131104
max_seq_len: 131104
tensor_parallel_size: 1
moe_expert_parallel_size: 1
enable_attention_dp: true
enable_attention_dp: false
pipeline_parallel_size: 8
print_iter_log: true
cuda_graph_config: null

4
tests/integration/defs/perf/disagg/test_configs/disagg/perf/deepseek-r1-fp4_128k8k_ctx5_pp4_gen1_dep16_bs16_eplb0_mtp0-Default.yaml
View File

@ -77,12 +77,12 @@ worker_config:
stream_interval: 20
num_postprocess_workers: 4
ctx:
max_batch_size: 16
max_batch_size: 1
max_num_tokens: 131104
max_seq_len: 131104
tensor_parallel_size: 1
moe_expert_parallel_size: 1
enable_attention_dp: true
enable_attention_dp: false
pipeline_parallel_size: 4
print_iter_log: true
cuda_graph_config: null

4
tests/integration/defs/perf/disagg/test_configs/disagg/perf/deepseek-r1-fp4_128k8k_ctx5_pp4_gen1_dep16_bs8_eplb0_mtp3-Default.yaml
View File

@ -81,12 +81,12 @@ worker_config:
decoding_type: MTP
num_nextn_predict_layers: 3
ctx:
max_batch_size: 8
max_batch_size: 1
max_num_tokens: 131104
max_seq_len: 131104
tensor_parallel_size: 1
moe_expert_parallel_size: 1
enable_attention_dp: true
enable_attention_dp: false
pipeline_parallel_size: 4
print_iter_log: true
cuda_graph_config: null

4
tests/integration/defs/perf/disagg/test_configs/disagg/perf/deepseek-r1-fp4_128k8k_ctx5_pp4_gen1_dep32_bs2_eplb0_mtp3-Default.yaml
View File

@ -81,12 +81,12 @@ worker_config:
decoding_type: MTP
num_nextn_predict_layers: 3
ctx:
max_batch_size: 2
max_batch_size: 1
max_num_tokens: 131104
max_seq_len: 131104
tensor_parallel_size: 1
moe_expert_parallel_size: 1
enable_attention_dp: true
enable_attention_dp: false
pipeline_parallel_size: 4
print_iter_log: true
cuda_graph_config: null

4
tests/integration/defs/perf/disagg/test_configs/disagg/perf/deepseek-r1-fp4_128k8k_ctx5_pp4_gen1_dep32_bs4_eplb0_mtp0-Default.yaml
View File

@ -77,12 +77,12 @@ worker_config:
stream_interval: 20
num_postprocess_workers: 4
ctx:
max_batch_size: 4
max_batch_size: 1
max_num_tokens: 131104
max_seq_len: 131104
tensor_parallel_size: 1
moe_expert_parallel_size: 1
enable_attention_dp: true
enable_attention_dp: false
pipeline_parallel_size: 4
print_iter_log: true
cuda_graph_config: null

4
tests/integration/defs/perf/disagg/test_configs/disagg/perf/deepseek-r1-fp4_128k8k_ctx7_pp4_gen1_dep16_bs16_eplb0_mtp1-Default.yaml
View File

@ -81,12 +81,12 @@ worker_config:
decoding_type: MTP
num_nextn_predict_layers: 1
ctx:
max_batch_size: 16
max_batch_size: 1
max_num_tokens: 131104
max_seq_len: 131104
tensor_parallel_size: 1
moe_expert_parallel_size: 1
enable_attention_dp: true
enable_attention_dp: false
pipeline_parallel_size: 4
print_iter_log: true
cuda_graph_config: null

4
tests/integration/defs/perf/disagg/test_configs/disagg/perf/deepseek-r1-fp4_128k8k_ctx7_pp4_gen1_dep16_bs32_eplb0_mtp0-Default.yaml
View File

@ -77,12 +77,12 @@ worker_config:
stream_interval: 20
num_postprocess_workers: 4
ctx:
max_batch_size: 32
max_batch_size: 1
max_num_tokens: 131104
max_seq_len: 131104
tensor_parallel_size: 1
moe_expert_parallel_size: 1
enable_attention_dp: true
enable_attention_dp: false
pipeline_parallel_size: 4
print_iter_log: true
cuda_graph_config: null

4
tests/integration/defs/perf/disagg/test_configs/disagg/perf/deepseek-r1-fp4_128k8k_ctx8_pp4_gen1_dep16_bs32_eplb0_mtp1-Default.yaml
View File

@ -81,12 +81,12 @@ worker_config:
decoding_type: MTP
num_nextn_predict_layers: 1
ctx:
max_batch_size: 32
max_batch_size: 1
max_num_tokens: 131104
max_seq_len: 131104
tensor_parallel_size: 1
moe_expert_parallel_size: 1
enable_attention_dp: true
enable_attention_dp: false
pipeline_parallel_size: 4
print_iter_log: true
cuda_graph_config: null

4
tests/integration/defs/perf/disagg/test_configs/disagg/perf/deepseek-r1-fp4_128k8k_ctx8_pp4_gen1_dep32_bs4_eplb0_mtp3-Default.yaml
View File

@ -81,12 +81,12 @@ worker_config:
decoding_type: MTP
num_nextn_predict_layers: 3
ctx:
max_batch_size: 4
max_batch_size: 1
max_num_tokens: 131104
max_seq_len: 131104
tensor_parallel_size: 1
moe_expert_parallel_size: 1
enable_attention_dp: true
enable_attention_dp: false
pipeline_parallel_size: 4
print_iter_log: true
cuda_graph_config: null

4
tests/integration/defs/perf/disagg/test_configs/disagg/perf/deepseek-r1-fp4_128k8k_ctx8_pp4_gen1_dep32_bs8_eplb0_mtp0-Default.yaml
View File

@ -77,12 +77,12 @@ worker_config:
stream_interval: 20
num_postprocess_workers: 4
ctx:
max_batch_size: 8
max_batch_size: 1
max_num_tokens: 131104
max_seq_len: 131104
tensor_parallel_size: 1
moe_expert_parallel_size: 1
enable_attention_dp: true
enable_attention_dp: false
pipeline_parallel_size: 4
print_iter_log: true
cuda_graph_config: null

4
tests/integration/defs/perf/disagg/test_configs/disagg/perf/deepseek-r1-fp4_128k8k_ctx8_pp4_gen1_dep32_bs8_eplb0_mtp3-Default.yaml
View File

@ -81,12 +81,12 @@ worker_config:
decoding_type: MTP
num_nextn_predict_layers: 3
ctx:
max_batch_size: 8
max_batch_size: 1
max_num_tokens: 131104
max_seq_len: 131104
tensor_parallel_size: 1
moe_expert_parallel_size: 1
enable_attention_dp: true
enable_attention_dp: false
pipeline_parallel_size: 4
print_iter_log: true
cuda_graph_config: null

1
tests/integration/defs/perf/test_perf.py
View File

@ -238,6 +238,7 @@ TRUST_REMOTE_CODE_MODELS = { # these models require explicit trust_remote_code=
"llama_v3.3_nemotron_super_49b_fp8",
"llama_v3.1_nemotron_ultra_253b",
"llama_v3.1_nemotron_ultra_253b_fp8",
"kimi_k2_nvfp4",
}
# Autodeploy model configs - maps model name to config file path (relative to TRT-LLM root)