Skip to content

parameter.md

Latest commit

 

History

History
93 lines (68 loc) · 2.69 KB

parameter.md

File metadata and controls

93 lines (68 loc) · 2.69 KB

Script Parameters Reference

scripts/start_search_service.sh

Start local search service for BrowseComp-Plus benchmark.

bash scripts/start_search_service.sh [searcher_type] [port] [cuda_visible_devices]
Parameter Default Description
searcher_type dense bm25 or dense
port 8000 Port for search service
cuda_visible_devices 0 GPU ID for dense searcher (ignored for BM25)

**Important!! You should make sure there is sufficient GPU memory for Dense search -- which would cost about 15GB

Examples:

# BM25 search (CPU-based, lightweight)
bash scripts/start_search_service.sh bm25 8000

# Dense search (GPU-based, better quality)
bash scripts/start_search_service.sh dense 8000 0

scripts/start_nemotron_servers.sh

Start multiple vLLM servers with tensor parallelism.

bash scripts/start_nemotron_servers.sh [tensor_parallel_size] [base_port] [cuda_visible_devices] [model_path]
Parameter Default Description
tensor_parallel_size 2 Number of GPUs per server
base_port 8001 Base port number for servers
cuda_visible_devices 0,1,2,3,4,5,6,7 Comma-separated GPU IDs to use
model_path (see script) Model name or path

Examples:

# 2 servers with TP=2 on 4 GPUs
bash scripts/start_nemotron_servers.sh 2 8001 0,1,2,3

# 4 servers with TP=2 on 8 GPUs
bash scripts/start_nemotron_servers.sh 2 8001 0,1,2,3,4,5,6,7

# 1 server with TP=4 on 4 GPUs
bash scripts/start_nemotron_servers.sh 4 8001 0,1,2,3

run_agent.sh

Main script for running the agent on benchmarks.

bash run_agent.sh [output_dir] [base_port] [num_servers] [dataset_name] [browser_backend] [model_path]
Parameter Default Description
output_dir results/browsecomp/output Output directory for results
base_port 8002 Base port number for vLLM servers
num_servers 3 Number of vLLM servers to use
dataset_name browsecomp Dataset key (see benchmarks.md)
browser_backend local local or serper
model_path OpenResearcher/OpenResearcher-30B-A3B Model name or path

Examples:

# BrowseComp-Plus with local search
bash run_agent.sh results/bc 8001 2 browsecomp_plus local OpenResearcher/OpenResearcher-30B-A3B

# GAIA with Serper API
bash run_agent.sh results/gaia 8001 2 gaia serper OpenResearcher/OpenResearcher-30B-A3B

scripts/stop_servers.sh

Stop all running vLLM servers.

bash scripts/stop_servers.sh

This script reads PIDs from logs/server_pids.txt and terminates all servers.