Skip to content

Configuration¶

Environment Variables¶

LLM Configuration¶

# Model path
export TRIAGE_LLM_MODEL=/path/to/llama-2-7b-chat.Q5_K_S.gguf

# Generation settings
export NLP_TRIAGE_LLM_GENERATOR=1
export NLP_TRIAGE_LLM_REWRITE_PROB=0.30
export NLP_TRIAGE_LLM_TEMPERATURE=0.2
export NLP_TRIAGE_LLM_MAX_RETRIES=3

# Debugging
export TRIAGE_LLM_DEBUG=1

Model Settings¶

# Context window
export TRIAGE_LLM_CTX=4096

# Max tokens for generation
export TRIAGE_LLM_MAX_TOKENS=512

# Temperature (0.0-1.0)
export TRIAGE_LLM_TEMP=0.2

CLI Configuration¶

Uncertainty Thresholds¶

# Default threshold
nlp-triage --threshold 0.50 "incident text"

# High confidence mode
nlp-triage --threshold 0.70 "incident text"

# Low confidence mode
nlp-triage --threshold 0.30 "incident text"

Difficulty Modes¶

default - Standard uncertainty handling
soc-medium - Moderate strictness
soc-hard - Maximum strictness for edge cases

Streamlit UI Configuration¶

The UI respects the same environment variables as the CLI.

Dataset Generation¶

Generation Parameters¶

# In generate_cyber_incidents.py
--n-events 50000           # Number of incidents
--chunk-size 1000          # Checkpointing frequency
--start-date 2024-01-01    # Date range start
--end-date 2024-12-31      # Date range end
--use-llm                  # Enable LLM rewriting
--rewrite-report audit.json # LLM statistics output

See Production Generation for details.