Skip to content

configuration.md

Latest commit

 

History

History
131 lines (108 loc) · 3.31 KB

configuration.md

File metadata and controls

131 lines (108 loc) · 3.31 KB
Raw

Configuration Reference

Distill-Align can be configured via a YAML or TOML file, or via environment variables.

Config File

Create a distill-align.yaml in your project root:

distill-align init

Full Example

# Project metadata
project:
  name: "my-dataset"
  version: "1.0"
  description: "Fine-tuning data for our domain"

# Ingestion configuration
ingestion:
  chunk_size: 1000
  chunk_overlap: 200
  respect_headers: true
  max_chunk_tokens: 4000
  sources:
    - path: ./data
      type: auto        # auto-detect or specify: markdown, code, pdf, docx, html, jupyter, json, csv, text
      recursive: true
      patterns: []      # Optional glob patterns

# Synthesis configuration
synthesis:
  provider: openai      # openai, ollama, vllm
  model: gpt-4o
  # base_url: http://localhost:11434  # For Ollama/vLLM
  # api_key: sk-...                    # Or use env var
  max_concurrency: 5
  max_rpm: 60
  temperature: 0.7
  socratic: true
  scaffold: true
  retry_attempts: 5

# Export configuration
export:
  formats:
    - sharegpt
    - alpaca
    - chatml
  output_dir: ./output
  generate_unsloth_script: true
  train_split: 0.9
  val_split: 0.05
  test_split: 0.05
  unsloth:
    model: unsloth/Meta-Llama-3.1-8B-Instruct
    max_seq_length: 2048
    lora_rank: 16
    lora_alpha: 16
    load_in_4bit: true
    batch_size: 2
    gradient_accumulation_steps: 4
    num_epochs: 3
    learning_rate: 0.0002

# Global settings
log_level: INFO
cache_dir: .cache

Environment Variables

All config options can be overridden via environment variables with the DISTILL_ prefix:

export DISTILL_LLM_PROVIDER=ollama
export DISTILL_LLM_MODEL=llama3.1
export DISTILL_LLM_BASE_URL=http://localhost:11434
export DISTILL_LLM_API_KEY=sk-...
export DISTILL_LOG_LEVEL=DEBUG
export DISTILL_CACHE_DIR=/var/cache/distill-align

API Key Environment Variables

Variable Required For Description
OPENAI_API_KEY OpenAI / Azure OpenAI API key
ANTHROPIC_API_KEY Anthropic Anthropic API key
GOOGLE_API_KEY Google Gemini Google AI Studio API key
AZURE_OPENAI_API_KEY Azure OpenAI Azure OpenAI resource key
AZURE_OPENAI_ENDPOINT Azure OpenAI Azure OpenAI endpoint URL
DISTILL_LLM_API_KEY Any provider Generic override (takes precedence)

CLI Override

CLI arguments override config file settings:

distill-align synthesize \
    --provider ollama \
    --model llama3.1 \
    --base-url http://localhost:11434 \
    --concurrency 3

Priority Order

Settings are applied in this order (highest priority first):

  1. CLI arguments
  2. Environment variables
  3. Config file (distill-align.yaml)
  4. Default values

Supported File Types

Extension Loader
.md, .markdown MarkdownLoader
.pdf PDFLoader
.py, .js, .ts, .java, .cpp, .go, .rs CodeLoader
.ipynb JupyterLoader
.docx DOCXLoader
.html, .htm HTMLLoader
.json, .jsonl JSONLoader
.csv CSVLoader
.txt, .log, .yaml, .toml TextLoader