kanshan/TensorRT-LLMs
1
0
Fork 0
mirror of https://github.com/NVIDIA/TensorRT-LLM.git synced 2026-01-26 13:43:38 +08:00
Code Issues Actions 1 Packages Projects Releases Wiki Activity
6095c80e56
TensorRT-LLMs/examples/configs/database/lookup.yaml
Venky fd1270b9ab
[TRTC-43] [feat] Add config db and docs (#9420) 
Signed-off-by: Frank Di Natale <3429989+FrankD412@users.noreply.github.com>
Signed-off-by: Venky Ganesh <23023424+venkywonka@users.noreply.github.com>
Co-authored-by: Frank Di Natale <3429989+FrankD412@users.noreply.github.com>
2025-12-12 04:00:03 +08:00

1177 lines
32 KiB
YAML
Raw Blame History

- model: nvidia/DeepSeek-R1-0528-FP4-v2
gpu: B200_NVL
isl: 1024
osl: 1024
concurrency: 128
config_path: examples/configs/database/nvidia/DeepSeek-R1-0528-FP4-v2/B200/1k1k_tp4_conc128.yaml
num_gpus: 4
- model: nvidia/DeepSeek-R1-0528-FP4-v2
gpu: B200_NVL
isl: 1024
osl: 1024
concurrency: 16
config_path: examples/configs/database/nvidia/DeepSeek-R1-0528-FP4-v2/B200/1k1k_tp4_conc16.yaml
num_gpus: 4
- model: nvidia/DeepSeek-R1-0528-FP4-v2
gpu: B200_NVL
isl: 1024
osl: 1024
concurrency: 256
config_path: examples/configs/database/nvidia/DeepSeek-R1-0528-FP4-v2/B200/1k1k_tp4_conc256.yaml
num_gpus: 4
- model: nvidia/DeepSeek-R1-0528-FP4-v2
gpu: B200_NVL
isl: 1024
osl: 1024
concurrency: 32
config_path: examples/configs/database/nvidia/DeepSeek-R1-0528-FP4-v2/B200/1k1k_tp4_conc32.yaml
num_gpus: 4
- model: nvidia/DeepSeek-R1-0528-FP4-v2
gpu: B200_NVL
isl: 1024
osl: 1024
concurrency: 4
config_path: examples/configs/database/nvidia/DeepSeek-R1-0528-FP4-v2/B200/1k1k_tp4_conc4.yaml
num_gpus: 4
- model: nvidia/DeepSeek-R1-0528-FP4-v2
gpu: B200_NVL
isl: 1024
osl: 1024
concurrency: 64
config_path: examples/configs/database/nvidia/DeepSeek-R1-0528-FP4-v2/B200/1k1k_tp4_conc64.yaml
num_gpus: 4
- model: nvidia/DeepSeek-R1-0528-FP4-v2
gpu: B200_NVL
isl: 1024
osl: 1024
concurrency: 8
config_path: examples/configs/database/nvidia/DeepSeek-R1-0528-FP4-v2/B200/1k1k_tp4_conc8.yaml
num_gpus: 4
- model: nvidia/DeepSeek-R1-0528-FP4-v2
gpu: B200_NVL
isl: 1024
osl: 1024
concurrency: 128
config_path: examples/configs/database/nvidia/DeepSeek-R1-0528-FP4-v2/B200/1k1k_tp8_conc128.yaml
num_gpus: 8
- model: nvidia/DeepSeek-R1-0528-FP4-v2
gpu: B200_NVL
isl: 1024
osl: 1024
concurrency: 16
config_path: examples/configs/database/nvidia/DeepSeek-R1-0528-FP4-v2/B200/1k1k_tp8_conc16.yaml
num_gpus: 8
- model: nvidia/DeepSeek-R1-0528-FP4-v2
gpu: B200_NVL
isl: 1024
osl: 1024
concurrency: 256
config_path: examples/configs/database/nvidia/DeepSeek-R1-0528-FP4-v2/B200/1k1k_tp8_conc256.yaml
num_gpus: 8
- model: nvidia/DeepSeek-R1-0528-FP4-v2
gpu: B200_NVL
isl: 1024
osl: 1024
concurrency: 32
config_path: examples/configs/database/nvidia/DeepSeek-R1-0528-FP4-v2/B200/1k1k_tp8_conc32.yaml
num_gpus: 8
- model: nvidia/DeepSeek-R1-0528-FP4-v2
gpu: B200_NVL
isl: 1024
osl: 1024
concurrency: 4
config_path: examples/configs/database/nvidia/DeepSeek-R1-0528-FP4-v2/B200/1k1k_tp8_conc4.yaml
num_gpus: 8
- model: nvidia/DeepSeek-R1-0528-FP4-v2
gpu: B200_NVL
isl: 1024
osl: 1024
concurrency: 64
config_path: examples/configs/database/nvidia/DeepSeek-R1-0528-FP4-v2/B200/1k1k_tp8_conc64.yaml
num_gpus: 8
- model: nvidia/DeepSeek-R1-0528-FP4-v2
gpu: B200_NVL
isl: 1024
osl: 1024
concurrency: 8
config_path: examples/configs/database/nvidia/DeepSeek-R1-0528-FP4-v2/B200/1k1k_tp8_conc8.yaml
num_gpus: 8
- model: nvidia/DeepSeek-R1-0528-FP4-v2
gpu: B200_NVL
isl: 8192
osl: 1024
concurrency: 128
config_path: examples/configs/database/nvidia/DeepSeek-R1-0528-FP4-v2/B200/8k1k_tp4_conc128.yaml
num_gpus: 4
- model: nvidia/DeepSeek-R1-0528-FP4-v2
gpu: B200_NVL
isl: 8192
osl: 1024
concurrency: 16
config_path: examples/configs/database/nvidia/DeepSeek-R1-0528-FP4-v2/B200/8k1k_tp4_conc16.yaml
num_gpus: 4
- model: nvidia/DeepSeek-R1-0528-FP4-v2
gpu: B200_NVL
isl: 8192
osl: 1024
concurrency: 256
config_path: examples/configs/database/nvidia/DeepSeek-R1-0528-FP4-v2/B200/8k1k_tp4_conc256.yaml
num_gpus: 4
- model: nvidia/DeepSeek-R1-0528-FP4-v2
gpu: B200_NVL
isl: 8192
osl: 1024
concurrency: 32
config_path: examples/configs/database/nvidia/DeepSeek-R1-0528-FP4-v2/B200/8k1k_tp4_conc32.yaml
num_gpus: 4
- model: nvidia/DeepSeek-R1-0528-FP4-v2
gpu: B200_NVL
isl: 8192
osl: 1024
concurrency: 4
config_path: examples/configs/database/nvidia/DeepSeek-R1-0528-FP4-v2/B200/8k1k_tp4_conc4.yaml
num_gpus: 4
- model: nvidia/DeepSeek-R1-0528-FP4-v2
gpu: B200_NVL
isl: 8192
osl: 1024
concurrency: 64
config_path: examples/configs/database/nvidia/DeepSeek-R1-0528-FP4-v2/B200/8k1k_tp4_conc64.yaml
num_gpus: 4
- model: nvidia/DeepSeek-R1-0528-FP4-v2
gpu: B200_NVL
isl: 8192
osl: 1024
concurrency: 8
config_path: examples/configs/database/nvidia/DeepSeek-R1-0528-FP4-v2/B200/8k1k_tp4_conc8.yaml
num_gpus: 4
- model: nvidia/DeepSeek-R1-0528-FP4-v2
gpu: B200_NVL
isl: 8192
osl: 1024
concurrency: 128
config_path: examples/configs/database/nvidia/DeepSeek-R1-0528-FP4-v2/B200/8k1k_tp8_conc128.yaml
num_gpus: 8
- model: nvidia/DeepSeek-R1-0528-FP4-v2
gpu: B200_NVL
isl: 8192
osl: 1024
concurrency: 16
config_path: examples/configs/database/nvidia/DeepSeek-R1-0528-FP4-v2/B200/8k1k_tp8_conc16.yaml
num_gpus: 8
- model: nvidia/DeepSeek-R1-0528-FP4-v2
gpu: B200_NVL
isl: 8192
osl: 1024
concurrency: 256
config_path: examples/configs/database/nvidia/DeepSeek-R1-0528-FP4-v2/B200/8k1k_tp8_conc256.yaml
num_gpus: 8
- model: nvidia/DeepSeek-R1-0528-FP4-v2
gpu: B200_NVL
isl: 8192
osl: 1024
concurrency: 32
config_path: examples/configs/database/nvidia/DeepSeek-R1-0528-FP4-v2/B200/8k1k_tp8_conc32.yaml
num_gpus: 8
- model: nvidia/DeepSeek-R1-0528-FP4-v2
gpu: B200_NVL
isl: 8192
osl: 1024
concurrency: 4
config_path: examples/configs/database/nvidia/DeepSeek-R1-0528-FP4-v2/B200/8k1k_tp8_conc4.yaml
num_gpus: 8
- model: nvidia/DeepSeek-R1-0528-FP4-v2
gpu: B200_NVL
isl: 8192
osl: 1024
concurrency: 64
config_path: examples/configs/database/nvidia/DeepSeek-R1-0528-FP4-v2/B200/8k1k_tp8_conc64.yaml
num_gpus: 8
- model: nvidia/DeepSeek-R1-0528-FP4-v2
gpu: B200_NVL
isl: 8192
osl: 1024
concurrency: 8
config_path: examples/configs/database/nvidia/DeepSeek-R1-0528-FP4-v2/B200/8k1k_tp8_conc8.yaml
num_gpus: 8
- model: deepseek-ai/DeepSeek-R1-0528
gpu: B200_NVL
isl: 1024
osl: 1024
concurrency: 16
config_path: examples/configs/database/deepseek-ai/DeepSeek-R1-0528/B200/1k1k_tp8_conc16.yaml
num_gpus: 8
- model: deepseek-ai/DeepSeek-R1-0528
gpu: B200_NVL
isl: 1024
osl: 1024
concurrency: 32
config_path: examples/configs/database/deepseek-ai/DeepSeek-R1-0528/B200/1k1k_tp8_conc32.yaml
num_gpus: 8
- model: deepseek-ai/DeepSeek-R1-0528
gpu: B200_NVL
isl: 1024
osl: 1024
concurrency: 4
config_path: examples/configs/database/deepseek-ai/DeepSeek-R1-0528/B200/1k1k_tp8_conc4.yaml
num_gpus: 8
- model: deepseek-ai/DeepSeek-R1-0528
gpu: B200_NVL
isl: 1024
osl: 1024
concurrency: 64
config_path: examples/configs/database/deepseek-ai/DeepSeek-R1-0528/B200/1k1k_tp8_conc64.yaml
num_gpus: 8
- model: deepseek-ai/DeepSeek-R1-0528
gpu: B200_NVL
isl: 1024
osl: 1024
concurrency: 8
config_path: examples/configs/database/deepseek-ai/DeepSeek-R1-0528/B200/1k1k_tp8_conc8.yaml
num_gpus: 8
- model: deepseek-ai/DeepSeek-R1-0528
gpu: B200_NVL
isl: 8192
osl: 1024
concurrency: 16
config_path: examples/configs/database/deepseek-ai/DeepSeek-R1-0528/B200/8k1k_tp8_conc16.yaml
num_gpus: 8
- model: deepseek-ai/DeepSeek-R1-0528
gpu: B200_NVL
isl: 8192
osl: 1024
concurrency: 32
config_path: examples/configs/database/deepseek-ai/DeepSeek-R1-0528/B200/8k1k_tp8_conc32.yaml
num_gpus: 8
- model: deepseek-ai/DeepSeek-R1-0528
gpu: B200_NVL
isl: 8192
osl: 1024
concurrency: 4
config_path: examples/configs/database/deepseek-ai/DeepSeek-R1-0528/B200/8k1k_tp8_conc4.yaml
num_gpus: 8
- model: deepseek-ai/DeepSeek-R1-0528
gpu: B200_NVL
isl: 8192
osl: 1024
concurrency: 64
config_path: examples/configs/database/deepseek-ai/DeepSeek-R1-0528/B200/8k1k_tp8_conc64.yaml
num_gpus: 8
- model: deepseek-ai/DeepSeek-R1-0528
gpu: B200_NVL
isl: 8192
osl: 1024
concurrency: 8
config_path: examples/configs/database/deepseek-ai/DeepSeek-R1-0528/B200/8k1k_tp8_conc8.yaml
num_gpus: 8
- model: deepseek-ai/DeepSeek-R1-0528
gpu: H200_SXM
isl: 1024
osl: 1024
concurrency: 16
config_path: examples/configs/database/deepseek-ai/DeepSeek-R1-0528/H200/1k1k_tp8_conc16.yaml
num_gpus: 8
- model: deepseek-ai/DeepSeek-R1-0528
gpu: H200_SXM
isl: 1024
osl: 1024
concurrency: 32
config_path: examples/configs/database/deepseek-ai/DeepSeek-R1-0528/H200/1k1k_tp8_conc32.yaml
num_gpus: 8
- model: deepseek-ai/DeepSeek-R1-0528
gpu: H200_SXM
isl: 1024
osl: 1024
concurrency: 4
config_path: examples/configs/database/deepseek-ai/DeepSeek-R1-0528/H200/1k1k_tp8_conc4.yaml
num_gpus: 8
- model: deepseek-ai/DeepSeek-R1-0528
gpu: H200_SXM
isl: 1024
osl: 1024
concurrency: 64
config_path: examples/configs/database/deepseek-ai/DeepSeek-R1-0528/H200/1k1k_tp8_conc64.yaml
num_gpus: 8
- model: deepseek-ai/DeepSeek-R1-0528
gpu: H200_SXM
isl: 1024
osl: 1024
concurrency: 8
config_path: examples/configs/database/deepseek-ai/DeepSeek-R1-0528/H200/1k1k_tp8_conc8.yaml
num_gpus: 8
- model: deepseek-ai/DeepSeek-R1-0528
gpu: H200_SXM
isl: 8192
osl: 1024
concurrency: 16
config_path: examples/configs/database/deepseek-ai/DeepSeek-R1-0528/H200/8k1k_tp8_conc16.yaml
num_gpus: 8
- model: deepseek-ai/DeepSeek-R1-0528
gpu: H200_SXM
isl: 8192
osl: 1024
concurrency: 32
config_path: examples/configs/database/deepseek-ai/DeepSeek-R1-0528/H200/8k1k_tp8_conc32.yaml
num_gpus: 8
- model: deepseek-ai/DeepSeek-R1-0528
gpu: H200_SXM
isl: 8192
osl: 1024
concurrency: 4
config_path: examples/configs/database/deepseek-ai/DeepSeek-R1-0528/H200/8k1k_tp8_conc4.yaml
num_gpus: 8
- model: deepseek-ai/DeepSeek-R1-0528
gpu: H200_SXM
isl: 8192
osl: 1024
concurrency: 64
config_path: examples/configs/database/deepseek-ai/DeepSeek-R1-0528/H200/8k1k_tp8_conc64.yaml
num_gpus: 8
- model: deepseek-ai/DeepSeek-R1-0528
gpu: H200_SXM
isl: 8192
osl: 1024
concurrency: 8
config_path: examples/configs/database/deepseek-ai/DeepSeek-R1-0528/H200/8k1k_tp8_conc8.yaml
num_gpus: 8
- model: openai/gpt-oss-120b
gpu: B200_NVL
isl: 1024
osl: 1024
concurrency: 16
config_path: examples/configs/database/openai/gpt-oss-120b/B200/1k1k_tp1_conc16.yaml
num_gpus: 1
- model: openai/gpt-oss-120b
gpu: B200_NVL
isl: 1024
osl: 1024
concurrency: 32
config_path: examples/configs/database/openai/gpt-oss-120b/B200/1k1k_tp1_conc32.yaml
num_gpus: 1
- model: openai/gpt-oss-120b
gpu: B200_NVL
isl: 1024
osl: 1024
concurrency: 4
config_path: examples/configs/database/openai/gpt-oss-120b/B200/1k1k_tp1_conc4.yaml
num_gpus: 1
- model: openai/gpt-oss-120b
gpu: B200_NVL
isl: 1024
osl: 1024
concurrency: 64
config_path: examples/configs/database/openai/gpt-oss-120b/B200/1k1k_tp1_conc64.yaml
num_gpus: 1
- model: openai/gpt-oss-120b
gpu: B200_NVL
isl: 1024
osl: 1024
concurrency: 8
config_path: examples/configs/database/openai/gpt-oss-120b/B200/1k1k_tp1_conc8.yaml
num_gpus: 1
- model: openai/gpt-oss-120b
gpu: B200_NVL
isl: 1024
osl: 1024
concurrency: 16
config_path: examples/configs/database/openai/gpt-oss-120b/B200/1k1k_tp2_conc16.yaml
num_gpus: 2
- model: openai/gpt-oss-120b
gpu: B200_NVL
isl: 1024
osl: 1024
concurrency: 32
config_path: examples/configs/database/openai/gpt-oss-120b/B200/1k1k_tp2_conc32.yaml
num_gpus: 2
- model: openai/gpt-oss-120b
gpu: B200_NVL
isl: 1024
osl: 1024
concurrency: 4
config_path: examples/configs/database/openai/gpt-oss-120b/B200/1k1k_tp2_conc4.yaml
num_gpus: 2
- model: openai/gpt-oss-120b
gpu: B200_NVL
isl: 1024
osl: 1024
concurrency: 64
config_path: examples/configs/database/openai/gpt-oss-120b/B200/1k1k_tp2_conc64.yaml
num_gpus: 2
- model: openai/gpt-oss-120b
gpu: B200_NVL
isl: 1024
osl: 1024
concurrency: 8
config_path: examples/configs/database/openai/gpt-oss-120b/B200/1k1k_tp2_conc8.yaml
num_gpus: 2
- model: openai/gpt-oss-120b
gpu: B200_NVL
isl: 1024
osl: 1024
concurrency: 16
config_path: examples/configs/database/openai/gpt-oss-120b/B200/1k1k_tp4_conc16.yaml
num_gpus: 4
- model: openai/gpt-oss-120b
gpu: B200_NVL
isl: 1024
osl: 1024
concurrency: 32
config_path: examples/configs/database/openai/gpt-oss-120b/B200/1k1k_tp4_conc32.yaml
num_gpus: 4
- model: openai/gpt-oss-120b
gpu: B200_NVL
isl: 1024
osl: 1024
concurrency: 4
config_path: examples/configs/database/openai/gpt-oss-120b/B200/1k1k_tp4_conc4.yaml
num_gpus: 4
- model: openai/gpt-oss-120b
gpu: B200_NVL
isl: 1024
osl: 1024
concurrency: 64
config_path: examples/configs/database/openai/gpt-oss-120b/B200/1k1k_tp4_conc64.yaml
num_gpus: 4
- model: openai/gpt-oss-120b
gpu: B200_NVL
isl: 1024
osl: 1024
concurrency: 8
config_path: examples/configs/database/openai/gpt-oss-120b/B200/1k1k_tp4_conc8.yaml
num_gpus: 4
- model: openai/gpt-oss-120b
gpu: B200_NVL
isl: 1024
osl: 1024
concurrency: 16
config_path: examples/configs/database/openai/gpt-oss-120b/B200/1k1k_tp8_conc16.yaml
num_gpus: 8
- model: openai/gpt-oss-120b
gpu: B200_NVL
isl: 1024
osl: 1024
concurrency: 32
config_path: examples/configs/database/openai/gpt-oss-120b/B200/1k1k_tp8_conc32.yaml
num_gpus: 8
- model: openai/gpt-oss-120b
gpu: B200_NVL
isl: 1024
osl: 1024
concurrency: 4
config_path: examples/configs/database/openai/gpt-oss-120b/B200/1k1k_tp8_conc4.yaml
num_gpus: 8
- model: openai/gpt-oss-120b
gpu: B200_NVL
isl: 1024
osl: 1024
concurrency: 64
config_path: examples/configs/database/openai/gpt-oss-120b/B200/1k1k_tp8_conc64.yaml
num_gpus: 8
- model: openai/gpt-oss-120b
gpu: B200_NVL
isl: 1024
osl: 1024
concurrency: 8
config_path: examples/configs/database/openai/gpt-oss-120b/B200/1k1k_tp8_conc8.yaml
num_gpus: 8
- model: openai/gpt-oss-120b
gpu: B200_NVL
isl: 1024
osl: 8192
concurrency: 16
config_path: examples/configs/database/openai/gpt-oss-120b/B200/1k8k_tp1_conc16.yaml
num_gpus: 1
- model: openai/gpt-oss-120b
gpu: B200_NVL
isl: 1024
osl: 8192
concurrency: 32
config_path: examples/configs/database/openai/gpt-oss-120b/B200/1k8k_tp1_conc32.yaml
num_gpus: 1
- model: openai/gpt-oss-120b
gpu: B200_NVL
isl: 1024
osl: 8192
concurrency: 4
config_path: examples/configs/database/openai/gpt-oss-120b/B200/1k8k_tp1_conc4.yaml
num_gpus: 1
- model: openai/gpt-oss-120b
gpu: B200_NVL
isl: 1024
osl: 8192
concurrency: 64
config_path: examples/configs/database/openai/gpt-oss-120b/B200/1k8k_tp1_conc64.yaml
num_gpus: 1
- model: openai/gpt-oss-120b
gpu: B200_NVL
isl: 1024
osl: 8192
concurrency: 8
config_path: examples/configs/database/openai/gpt-oss-120b/B200/1k8k_tp1_conc8.yaml
num_gpus: 1
- model: openai/gpt-oss-120b
gpu: B200_NVL
isl: 1024
osl: 8192
concurrency: 16
config_path: examples/configs/database/openai/gpt-oss-120b/B200/1k8k_tp2_conc16.yaml
num_gpus: 2
- model: openai/gpt-oss-120b
gpu: B200_NVL
isl: 1024
osl: 8192
concurrency: 32
config_path: examples/configs/database/openai/gpt-oss-120b/B200/1k8k_tp2_conc32.yaml
num_gpus: 2
- model: openai/gpt-oss-120b
gpu: B200_NVL
isl: 1024
osl: 8192
concurrency: 4
config_path: examples/configs/database/openai/gpt-oss-120b/B200/1k8k_tp2_conc4.yaml
num_gpus: 2
- model: openai/gpt-oss-120b
gpu: B200_NVL
isl: 1024
osl: 8192
concurrency: 64
config_path: examples/configs/database/openai/gpt-oss-120b/B200/1k8k_tp2_conc64.yaml
num_gpus: 2
- model: openai/gpt-oss-120b
gpu: B200_NVL
isl: 1024
osl: 8192
concurrency: 8
config_path: examples/configs/database/openai/gpt-oss-120b/B200/1k8k_tp2_conc8.yaml
num_gpus: 2
- model: openai/gpt-oss-120b
gpu: B200_NVL
isl: 1024
osl: 8192
concurrency: 16
config_path: examples/configs/database/openai/gpt-oss-120b/B200/1k8k_tp4_conc16.yaml
num_gpus: 4
- model: openai/gpt-oss-120b
gpu: B200_NVL
isl: 1024
osl: 8192
concurrency: 32
config_path: examples/configs/database/openai/gpt-oss-120b/B200/1k8k_tp4_conc32.yaml
num_gpus: 4
- model: openai/gpt-oss-120b
gpu: B200_NVL
isl: 1024
osl: 8192
concurrency: 4
config_path: examples/configs/database/openai/gpt-oss-120b/B200/1k8k_tp4_conc4.yaml
num_gpus: 4
- model: openai/gpt-oss-120b
gpu: B200_NVL
isl: 1024
osl: 8192
concurrency: 64
config_path: examples/configs/database/openai/gpt-oss-120b/B200/1k8k_tp4_conc64.yaml
num_gpus: 4
- model: openai/gpt-oss-120b
gpu: B200_NVL
isl: 1024
osl: 8192
concurrency: 8
config_path: examples/configs/database/openai/gpt-oss-120b/B200/1k8k_tp4_conc8.yaml
num_gpus: 4
- model: openai/gpt-oss-120b
gpu: B200_NVL
isl: 1024
osl: 8192
concurrency: 16
config_path: examples/configs/database/openai/gpt-oss-120b/B200/1k8k_tp8_conc16.yaml
num_gpus: 8
- model: openai/gpt-oss-120b
gpu: B200_NVL
isl: 1024
osl: 8192
concurrency: 32
config_path: examples/configs/database/openai/gpt-oss-120b/B200/1k8k_tp8_conc32.yaml
num_gpus: 8
- model: openai/gpt-oss-120b
gpu: B200_NVL
isl: 1024
osl: 8192
concurrency: 4
config_path: examples/configs/database/openai/gpt-oss-120b/B200/1k8k_tp8_conc4.yaml
num_gpus: 8
- model: openai/gpt-oss-120b
gpu: B200_NVL
isl: 1024
osl: 8192
concurrency: 64
config_path: examples/configs/database/openai/gpt-oss-120b/B200/1k8k_tp8_conc64.yaml
num_gpus: 8
- model: openai/gpt-oss-120b
gpu: B200_NVL
isl: 1024
osl: 8192
concurrency: 8
config_path: examples/configs/database/openai/gpt-oss-120b/B200/1k8k_tp8_conc8.yaml
num_gpus: 8
- model: openai/gpt-oss-120b
gpu: B200_NVL
isl: 8192
osl: 1024
concurrency: 16
config_path: examples/configs/database/openai/gpt-oss-120b/B200/8k1k_tp1_conc16.yaml
num_gpus: 1
- model: openai/gpt-oss-120b
gpu: B200_NVL
isl: 8192
osl: 1024
concurrency: 32
config_path: examples/configs/database/openai/gpt-oss-120b/B200/8k1k_tp1_conc32.yaml
num_gpus: 1
- model: openai/gpt-oss-120b
gpu: B200_NVL
isl: 8192
osl: 1024
concurrency: 4
config_path: examples/configs/database/openai/gpt-oss-120b/B200/8k1k_tp1_conc4.yaml
num_gpus: 1
- model: openai/gpt-oss-120b
gpu: B200_NVL
isl: 8192
osl: 1024
concurrency: 64
config_path: examples/configs/database/openai/gpt-oss-120b/B200/8k1k_tp1_conc64.yaml
num_gpus: 1
- model: openai/gpt-oss-120b
gpu: B200_NVL
isl: 8192
osl: 1024
concurrency: 8
config_path: examples/configs/database/openai/gpt-oss-120b/B200/8k1k_tp1_conc8.yaml
num_gpus: 1
- model: openai/gpt-oss-120b
gpu: B200_NVL
isl: 8192
osl: 1024
concurrency: 16
config_path: examples/configs/database/openai/gpt-oss-120b/B200/8k1k_tp2_conc16.yaml
num_gpus: 2
- model: openai/gpt-oss-120b
gpu: B200_NVL
isl: 8192
osl: 1024
concurrency: 32
config_path: examples/configs/database/openai/gpt-oss-120b/B200/8k1k_tp2_conc32.yaml
num_gpus: 2
- model: openai/gpt-oss-120b
gpu: B200_NVL
isl: 8192
osl: 1024
concurrency: 4
config_path: examples/configs/database/openai/gpt-oss-120b/B200/8k1k_tp2_conc4.yaml
num_gpus: 2
- model: openai/gpt-oss-120b
gpu: B200_NVL
isl: 8192
osl: 1024
concurrency: 64
config_path: examples/configs/database/openai/gpt-oss-120b/B200/8k1k_tp2_conc64.yaml
num_gpus: 2
- model: openai/gpt-oss-120b
gpu: B200_NVL
isl: 8192
osl: 1024
concurrency: 8
config_path: examples/configs/database/openai/gpt-oss-120b/B200/8k1k_tp2_conc8.yaml
num_gpus: 2
- model: openai/gpt-oss-120b
gpu: B200_NVL
isl: 8192
osl: 1024
concurrency: 16
config_path: examples/configs/database/openai/gpt-oss-120b/B200/8k1k_tp4_conc16.yaml
num_gpus: 4
- model: openai/gpt-oss-120b
gpu: B200_NVL
isl: 8192
osl: 1024
concurrency: 32
config_path: examples/configs/database/openai/gpt-oss-120b/B200/8k1k_tp4_conc32.yaml
num_gpus: 4
- model: openai/gpt-oss-120b
gpu: B200_NVL
isl: 8192
osl: 1024
concurrency: 4
config_path: examples/configs/database/openai/gpt-oss-120b/B200/8k1k_tp4_conc4.yaml
num_gpus: 4
- model: openai/gpt-oss-120b
gpu: B200_NVL
isl: 8192
osl: 1024
concurrency: 64
config_path: examples/configs/database/openai/gpt-oss-120b/B200/8k1k_tp4_conc64.yaml
num_gpus: 4
- model: openai/gpt-oss-120b
gpu: B200_NVL
isl: 8192
osl: 1024
concurrency: 8
config_path: examples/configs/database/openai/gpt-oss-120b/B200/8k1k_tp4_conc8.yaml
num_gpus: 4
- model: openai/gpt-oss-120b
gpu: B200_NVL
isl: 8192
osl: 1024
concurrency: 16
config_path: examples/configs/database/openai/gpt-oss-120b/B200/8k1k_tp8_conc16.yaml
num_gpus: 8
- model: openai/gpt-oss-120b
gpu: B200_NVL
isl: 8192
osl: 1024
concurrency: 32
config_path: examples/configs/database/openai/gpt-oss-120b/B200/8k1k_tp8_conc32.yaml
num_gpus: 8
- model: openai/gpt-oss-120b
gpu: B200_NVL
isl: 8192
osl: 1024
concurrency: 4
config_path: examples/configs/database/openai/gpt-oss-120b/B200/8k1k_tp8_conc4.yaml
num_gpus: 8
- model: openai/gpt-oss-120b
gpu: B200_NVL
isl: 8192
osl: 1024
concurrency: 64
config_path: examples/configs/database/openai/gpt-oss-120b/B200/8k1k_tp8_conc64.yaml
num_gpus: 8
- model: openai/gpt-oss-120b
gpu: B200_NVL
isl: 8192
osl: 1024
concurrency: 8
config_path: examples/configs/database/openai/gpt-oss-120b/B200/8k1k_tp8_conc8.yaml
num_gpus: 8
- model: openai/gpt-oss-120b
gpu: H200_SXM
isl: 1024
osl: 1024
concurrency: 16
config_path: examples/configs/database/openai/gpt-oss-120b/H200/1k1k_tp1_conc16.yaml
num_gpus: 1
- model: openai/gpt-oss-120b
gpu: H200_SXM
isl: 1024
osl: 1024
concurrency: 32
config_path: examples/configs/database/openai/gpt-oss-120b/H200/1k1k_tp1_conc32.yaml
num_gpus: 1
- model: openai/gpt-oss-120b
gpu: H200_SXM
isl: 1024
osl: 1024
concurrency: 4
config_path: examples/configs/database/openai/gpt-oss-120b/H200/1k1k_tp1_conc4.yaml
num_gpus: 1
- model: openai/gpt-oss-120b
gpu: H200_SXM
isl: 1024
osl: 1024
concurrency: 64
config_path: examples/configs/database/openai/gpt-oss-120b/H200/1k1k_tp1_conc64.yaml
num_gpus: 1
- model: openai/gpt-oss-120b
gpu: H200_SXM
isl: 1024
osl: 1024
concurrency: 8
config_path: examples/configs/database/openai/gpt-oss-120b/H200/1k1k_tp1_conc8.yaml
num_gpus: 1
- model: openai/gpt-oss-120b
gpu: H200_SXM
isl: 1024
osl: 1024
concurrency: 16
config_path: examples/configs/database/openai/gpt-oss-120b/H200/1k1k_tp2_conc16.yaml
num_gpus: 2
- model: openai/gpt-oss-120b
gpu: H200_SXM
isl: 1024
osl: 1024
concurrency: 32
config_path: examples/configs/database/openai/gpt-oss-120b/H200/1k1k_tp2_conc32.yaml
num_gpus: 2
- model: openai/gpt-oss-120b
gpu: H200_SXM
isl: 1024
osl: 1024
concurrency: 4
config_path: examples/configs/database/openai/gpt-oss-120b/H200/1k1k_tp2_conc4.yaml
num_gpus: 2
- model: openai/gpt-oss-120b
gpu: H200_SXM
isl: 1024
osl: 1024
concurrency: 64
config_path: examples/configs/database/openai/gpt-oss-120b/H200/1k1k_tp2_conc64.yaml
num_gpus: 2
- model: openai/gpt-oss-120b
gpu: H200_SXM
isl: 1024
osl: 1024
concurrency: 8
config_path: examples/configs/database/openai/gpt-oss-120b/H200/1k1k_tp2_conc8.yaml
num_gpus: 2
- model: openai/gpt-oss-120b
gpu: H200_SXM
isl: 1024
osl: 1024
concurrency: 16
config_path: examples/configs/database/openai/gpt-oss-120b/H200/1k1k_tp4_conc16.yaml
num_gpus: 4
- model: openai/gpt-oss-120b
gpu: H200_SXM
isl: 1024
osl: 1024
concurrency: 32
config_path: examples/configs/database/openai/gpt-oss-120b/H200/1k1k_tp4_conc32.yaml
num_gpus: 4
- model: openai/gpt-oss-120b
gpu: H200_SXM
isl: 1024
osl: 1024
concurrency: 4
config_path: examples/configs/database/openai/gpt-oss-120b/H200/1k1k_tp4_conc4.yaml
num_gpus: 4
- model: openai/gpt-oss-120b
gpu: H200_SXM
isl: 1024
osl: 1024
concurrency: 64
config_path: examples/configs/database/openai/gpt-oss-120b/H200/1k1k_tp4_conc64.yaml
num_gpus: 4
- model: openai/gpt-oss-120b
gpu: H200_SXM
isl: 1024
osl: 1024
concurrency: 8
config_path: examples/configs/database/openai/gpt-oss-120b/H200/1k1k_tp4_conc8.yaml
num_gpus: 4
- model: openai/gpt-oss-120b
gpu: H200_SXM
isl: 1024
osl: 1024
concurrency: 16
config_path: examples/configs/database/openai/gpt-oss-120b/H200/1k1k_tp8_conc16.yaml
num_gpus: 8
- model: openai/gpt-oss-120b
gpu: H200_SXM
isl: 1024
osl: 1024
concurrency: 32
config_path: examples/configs/database/openai/gpt-oss-120b/H200/1k1k_tp8_conc32.yaml
num_gpus: 8
- model: openai/gpt-oss-120b
gpu: H200_SXM
isl: 1024
osl: 1024
concurrency: 4
config_path: examples/configs/database/openai/gpt-oss-120b/H200/1k1k_tp8_conc4.yaml
num_gpus: 8
- model: openai/gpt-oss-120b
gpu: H200_SXM
isl: 1024
osl: 1024
concurrency: 64
config_path: examples/configs/database/openai/gpt-oss-120b/H200/1k1k_tp8_conc64.yaml
num_gpus: 8
- model: openai/gpt-oss-120b
gpu: H200_SXM
isl: 1024
osl: 1024
concurrency: 8
config_path: examples/configs/database/openai/gpt-oss-120b/H200/1k1k_tp8_conc8.yaml
num_gpus: 8
- model: openai/gpt-oss-120b
gpu: H200_SXM
isl: 1024
osl: 8192
concurrency: 16
config_path: examples/configs/database/openai/gpt-oss-120b/H200/1k8k_tp1_conc16.yaml
num_gpus: 1
- model: openai/gpt-oss-120b
gpu: H200_SXM
isl: 1024
osl: 8192
concurrency: 32
config_path: examples/configs/database/openai/gpt-oss-120b/H200/1k8k_tp1_conc32.yaml
num_gpus: 1
- model: openai/gpt-oss-120b
gpu: H200_SXM
isl: 1024
osl: 8192
concurrency: 4
config_path: examples/configs/database/openai/gpt-oss-120b/H200/1k8k_tp1_conc4.yaml
num_gpus: 1
- model: openai/gpt-oss-120b
gpu: H200_SXM
isl: 1024
osl: 8192
concurrency: 64
config_path: examples/configs/database/openai/gpt-oss-120b/H200/1k8k_tp1_conc64.yaml
num_gpus: 1
- model: openai/gpt-oss-120b
gpu: H200_SXM
isl: 1024
osl: 8192
concurrency: 8
config_path: examples/configs/database/openai/gpt-oss-120b/H200/1k8k_tp1_conc8.yaml
num_gpus: 1
- model: openai/gpt-oss-120b
gpu: H200_SXM
isl: 1024
osl: 8192
concurrency: 16
config_path: examples/configs/database/openai/gpt-oss-120b/H200/1k8k_tp2_conc16.yaml
num_gpus: 2
- model: openai/gpt-oss-120b
gpu: H200_SXM
isl: 1024
osl: 8192
concurrency: 32
config_path: examples/configs/database/openai/gpt-oss-120b/H200/1k8k_tp2_conc32.yaml
num_gpus: 2
- model: openai/gpt-oss-120b
gpu: H200_SXM
isl: 1024
osl: 8192
concurrency: 4
config_path: examples/configs/database/openai/gpt-oss-120b/H200/1k8k_tp2_conc4.yaml
num_gpus: 2
- model: openai/gpt-oss-120b
gpu: H200_SXM
isl: 1024
osl: 8192
concurrency: 64
config_path: examples/configs/database/openai/gpt-oss-120b/H200/1k8k_tp2_conc64.yaml
num_gpus: 2
- model: openai/gpt-oss-120b
gpu: H200_SXM
isl: 1024
osl: 8192
concurrency: 8
config_path: examples/configs/database/openai/gpt-oss-120b/H200/1k8k_tp2_conc8.yaml
num_gpus: 2
- model: openai/gpt-oss-120b
gpu: H200_SXM
isl: 1024
osl: 8192
concurrency: 16
config_path: examples/configs/database/openai/gpt-oss-120b/H200/1k8k_tp4_conc16.yaml
num_gpus: 4
- model: openai/gpt-oss-120b
gpu: H200_SXM
isl: 1024
osl: 8192
concurrency: 32
config_path: examples/configs/database/openai/gpt-oss-120b/H200/1k8k_tp4_conc32.yaml
num_gpus: 4
- model: openai/gpt-oss-120b
gpu: H200_SXM
isl: 1024
osl: 8192
concurrency: 4
config_path: examples/configs/database/openai/gpt-oss-120b/H200/1k8k_tp4_conc4.yaml
num_gpus: 4
- model: openai/gpt-oss-120b
gpu: H200_SXM
isl: 1024
osl: 8192
concurrency: 64
config_path: examples/configs/database/openai/gpt-oss-120b/H200/1k8k_tp4_conc64.yaml
num_gpus: 4
- model: openai/gpt-oss-120b
gpu: H200_SXM
isl: 1024
osl: 8192
concurrency: 8
config_path: examples/configs/database/openai/gpt-oss-120b/H200/1k8k_tp4_conc8.yaml
num_gpus: 4
- model: openai/gpt-oss-120b
gpu: H200_SXM
isl: 1024
osl: 8192
concurrency: 16
config_path: examples/configs/database/openai/gpt-oss-120b/H200/1k8k_tp8_conc16.yaml
num_gpus: 8
- model: openai/gpt-oss-120b
gpu: H200_SXM
isl: 1024
osl: 8192
concurrency: 32
config_path: examples/configs/database/openai/gpt-oss-120b/H200/1k8k_tp8_conc32.yaml
num_gpus: 8
- model: openai/gpt-oss-120b
gpu: H200_SXM
isl: 1024
osl: 8192
concurrency: 4
config_path: examples/configs/database/openai/gpt-oss-120b/H200/1k8k_tp8_conc4.yaml
num_gpus: 8
- model: openai/gpt-oss-120b
gpu: H200_SXM
isl: 1024
osl: 8192
concurrency: 64
config_path: examples/configs/database/openai/gpt-oss-120b/H200/1k8k_tp8_conc64.yaml
num_gpus: 8
- model: openai/gpt-oss-120b
gpu: H200_SXM
isl: 1024
osl: 8192
concurrency: 8
config_path: examples/configs/database/openai/gpt-oss-120b/H200/1k8k_tp8_conc8.yaml
num_gpus: 8
- model: openai/gpt-oss-120b
gpu: H200_SXM
isl: 8192
osl: 1024
concurrency: 16
config_path: examples/configs/database/openai/gpt-oss-120b/H200/8k1k_tp1_conc16.yaml
num_gpus: 1
- model: openai/gpt-oss-120b
gpu: H200_SXM
isl: 8192
osl: 1024
concurrency: 32
config_path: examples/configs/database/openai/gpt-oss-120b/H200/8k1k_tp1_conc32.yaml
num_gpus: 1
- model: openai/gpt-oss-120b
gpu: H200_SXM
isl: 8192
osl: 1024
concurrency: 4
config_path: examples/configs/database/openai/gpt-oss-120b/H200/8k1k_tp1_conc4.yaml
num_gpus: 1
- model: openai/gpt-oss-120b
gpu: H200_SXM
isl: 8192
osl: 1024
concurrency: 64
config_path: examples/configs/database/openai/gpt-oss-120b/H200/8k1k_tp1_conc64.yaml
num_gpus: 1
- model: openai/gpt-oss-120b
gpu: H200_SXM
isl: 8192
osl: 1024
concurrency: 8
config_path: examples/configs/database/openai/gpt-oss-120b/H200/8k1k_tp1_conc8.yaml
num_gpus: 1
- model: openai/gpt-oss-120b
gpu: H200_SXM
isl: 8192
osl: 1024
concurrency: 16
config_path: examples/configs/database/openai/gpt-oss-120b/H200/8k1k_tp2_conc16.yaml
num_gpus: 2
- model: openai/gpt-oss-120b
gpu: H200_SXM
isl: 8192
osl: 1024
concurrency: 32
config_path: examples/configs/database/openai/gpt-oss-120b/H200/8k1k_tp2_conc32.yaml
num_gpus: 2
- model: openai/gpt-oss-120b
gpu: H200_SXM
isl: 8192
osl: 1024
concurrency: 4
config_path: examples/configs/database/openai/gpt-oss-120b/H200/8k1k_tp2_conc4.yaml
num_gpus: 2
- model: openai/gpt-oss-120b
gpu: H200_SXM
isl: 8192
osl: 1024
concurrency: 64
config_path: examples/configs/database/openai/gpt-oss-120b/H200/8k1k_tp2_conc64.yaml
num_gpus: 2
- model: openai/gpt-oss-120b
gpu: H200_SXM
isl: 8192
osl: 1024
concurrency: 8
config_path: examples/configs/database/openai/gpt-oss-120b/H200/8k1k_tp2_conc8.yaml
num_gpus: 2
- model: openai/gpt-oss-120b
gpu: H200_SXM
isl: 8192
osl: 1024
concurrency: 16
config_path: examples/configs/database/openai/gpt-oss-120b/H200/8k1k_tp4_conc16.yaml
num_gpus: 4
- model: openai/gpt-oss-120b
gpu: H200_SXM
isl: 8192
osl: 1024
concurrency: 32
config_path: examples/configs/database/openai/gpt-oss-120b/H200/8k1k_tp4_conc32.yaml
num_gpus: 4
- model: openai/gpt-oss-120b
gpu: H200_SXM
isl: 8192
osl: 1024
concurrency: 4
config_path: examples/configs/database/openai/gpt-oss-120b/H200/8k1k_tp4_conc4.yaml
num_gpus: 4
- model: openai/gpt-oss-120b
gpu: H200_SXM
isl: 8192
osl: 1024
concurrency: 64
config_path: examples/configs/database/openai/gpt-oss-120b/H200/8k1k_tp4_conc64.yaml
num_gpus: 4
- model: openai/gpt-oss-120b
gpu: H200_SXM
isl: 8192
osl: 1024
concurrency: 8
config_path: examples/configs/database/openai/gpt-oss-120b/H200/8k1k_tp4_conc8.yaml
num_gpus: 4
- model: openai/gpt-oss-120b
gpu: H200_SXM
isl: 8192
osl: 1024
concurrency: 16
config_path: examples/configs/database/openai/gpt-oss-120b/H200/8k1k_tp8_conc16.yaml
num_gpus: 8
- model: openai/gpt-oss-120b
gpu: H200_SXM
isl: 8192
osl: 1024
concurrency: 32
config_path: examples/configs/database/openai/gpt-oss-120b/H200/8k1k_tp8_conc32.yaml
num_gpus: 8
- model: openai/gpt-oss-120b
gpu: H200_SXM
isl: 8192
osl: 1024
concurrency: 4
config_path: examples/configs/database/openai/gpt-oss-120b/H200/8k1k_tp8_conc4.yaml
num_gpus: 8
- model: openai/gpt-oss-120b
gpu: H200_SXM
isl: 8192
osl: 1024
concurrency: 64
config_path: examples/configs/database/openai/gpt-oss-120b/H200/8k1k_tp8_conc64.yaml
num_gpus: 8
- model: openai/gpt-oss-120b
gpu: H200_SXM
isl: 8192
osl: 1024
concurrency: 8
config_path: examples/configs/database/openai/gpt-oss-120b/H200/8k1k_tp8_conc8.yaml
num_gpus: 8
Reference in New Issue View Git Blame Copy Permalink