TensorRT-LLMs/examples/wide_ep/slurm_scripts/config.yaml
Kaiyu Xie 0788635d6c
[TRTLLM-9762] [doc] Update documents for GB300 NVL72 (#9987)
Signed-off-by: Kaiyu Xie <26294424+kaiyux@users.noreply.github.com>
2025-12-14 19:30:28 -08:00

115 lines
3.6 KiB
YAML

# SLURM Configuration
slurm:
script_file: "disaggr_torch.slurm"
partition: "<partition>"
account: "<account>"
job_time: "02:00:00"
job_name: "<job_name>"
extra_args: "" # Cluster specific arguments, e.g. "--gres=gpu:4 --exclude=node1,node2"
numa_bind: true # Only enable for GB200/GB300 NVL72
# Benchmark Mode
benchmark:
mode: "e2e" # Options: e2e, gen_only
use_nv_sa_benchmark: false # Whether to use NVIDIA SA benchmark script
multi_round: 8 # Number of benchmark rounds
benchmark_ratio: 0.8 # Benchmark ratio
streaming: true # Enable streaming mode
concurrency_list: "1024"
input_length: 8196 # Input sequence length
output_length: 1024 # Output sequence length
dataset_file: "<dataset_file>"
# Hardware Configuration
hardware:
gpus_per_node: 4 # Modify this with your hardware configuration
num_ctx_servers: 1 # Number of context servers
num_gen_servers: 1 # Number of generation servers
# Environment Configuration
environment:
container_mount: "<container_mount>" # Format: path1:path1,path2:path2
container_image: "<container_image>"
model_path: "<model_path>"
trtllm_repo: "<trtllm_repo>"
build_wheel: false # Don't build the wheel when launching multiple jobs
trtllm_wheel_path: "" # Path to pre-built TensorRT-LLM wheel. If provided, install from this wheel instead
work_dir: "<full_path_to_work_dir>"
worker_env_var: "TLLM_LOG_LEVEL=INFO TRTLLM_SERVER_DISABLE_GC=1 TRTLLM_WORKER_DISABLE_GC=1 TRTLLM_ENABLE_PDL=1 ENROOT_ALLOW_DEV=yes"
server_env_var: "TRTLLM_SERVER_DISABLE_GC=1"
# Profiling Configuration
profiling:
nsys_on: false # Set to true to enable profiling
ctx_profile_range: "10-30" # Set TLLM_PROFILE_START_STOP for ctx workers
gen_profile_range: "200-250" # Set TLLM_PROFILE_START_STOP for gen workers
# Accuracy Configuration
accuracy:
enable_accuracy_test: false # Set to true to enable accuracy evaluation
model: "local-completions" # Model type for lm_eval
tasks: "gsm8k" # Evaluation tasks (comma-separated)
model_args_extra: "num_concurrent=512,max_retries=3,tokenized_requests=false,timeout=1200,max_gen_toks=256,max_length=4096" # Extra model arguments for lm_eval
worker_config:
gen:
tensor_parallel_size: 32
moe_expert_parallel_size: 32
enable_attention_dp: true
enable_lm_head_tp_in_adp: true
pipeline_parallel_size: 1
max_batch_size: 128
max_num_tokens: 512
max_seq_len: 9236
cuda_graph_config:
enable_padding: true
batch_sizes:
- 1
- 2
- 4
- 8
- 16
- 32
- 64
- 128
print_iter_log: true
kv_cache_config:
enable_block_reuse: false
free_gpu_memory_fraction: 0.6
dtype: fp8
moe_config:
backend: WIDEEP
use_low_precision_moe_combine: true
load_balancer:
num_slots: 288
layer_updates_per_iter: 1
cache_transceiver_config:
max_tokens_in_buffer: 8448
backend: DEFAULT
stream_interval: 20
num_postprocess_workers: 4
speculative_config:
decoding_type: MTP
num_nextn_predict_layers: 3
ctx:
max_batch_size: 1
max_num_tokens: 8448
max_seq_len: 8212
tensor_parallel_size: 4
moe_expert_parallel_size: 4
enable_attention_dp: true
pipeline_parallel_size: 1
print_iter_log: true
cuda_graph_config: null
disable_overlap_scheduler: true
kv_cache_config:
enable_block_reuse: false
free_gpu_memory_fraction: 0.75
dtype: fp8
cache_transceiver_config:
max_tokens_in_buffer: 8448
backend: DEFAULT
speculative_config:
decoding_type: MTP
num_nextn_predict_layers: 3