Skip to content

CLI Reference

Fuse provides a command-line interface for extraction, training, and quantization.

fuse --help

fuse extract

Run structured extraction on text input.

With a config file

fuse extract "Sarah Chen is a 34-year-old architect at Stripe" \
  --config extract_person.yaml

With inline flags

# HuggingFace model — auto-downloads
fuse extract "SpaceX was founded in 2002" \
  --model bartowski/Phi-4-mini-instruct-GGUF \
  --fields "company:str,year:int,industry:str"

# Local GGUF model
fuse extract "John is 30" \
  --model ./model.gguf \
  --fields "name:str,age:int"

# Using a JSON schema file
fuse extract "John is 30 and knows Python" \
  --model bartowski/Llama-3.2-1B-Instruct-GGUF \
  --schema schema.json

With uvx (no install)

uvx fusellm extract "Sarah Chen is a 34-year-old architect" \
  --model bartowski/Llama-3.2-1B-Instruct-GGUF \
  --fields "name:str,age:int,job_title:str"

With evidence spans

Add --spans to see where each extracted value was found in the source text:

fuse extract "Sarah Chen is a 34-year-old architect at Stripe" \
  --model bartowski/Llama-3.2-1B-Instruct-GGUF \
  --fields "name:str,age:int,job_title:str,company:str" \
  --spans

Output includes evidence quotes, explicit/implicit classification, and character offsets.

HTML visualization

Generate an interactive HTML file with color-coded highlighted spans:

fuse extract "Sarah Chen is a 34-year-old architect at Stripe" \
  --model bartowski/Llama-3.2-1B-Instruct-GGUF \
  --fields "name:str,age:int,job_title:str,company:str" \
  --html result.html

The HTML output shows:

  • Each field highlighted with a distinct color in the source text
  • Solid outlines for explicit extractions (value is verbatim in text)
  • Dashed outlines for implicit extractions (value is inferred from context)
  • Hover tooltips with field name and value
  • A legend with field details, types, and character offsets

Tip

--html implies --spans — you don't need to pass both.

Options

Option Description
--config PATH YAML config file for extraction
--model TEXT Model path (local GGUF) or HuggingFace repo name
--fields TEXT Comma-separated field definitions (e.g., name:str,age:int)
--schema PATH JSON schema file path
--format TEXT Prompt format: llama, chatml, or generic (default: llama)
--max-tokens INT Maximum tokens to generate (default: 512)
--spans Include evidence spans with source text localization
--html PATH Write highlighted HTML visualization to file

fuse train

Fine-tune a model using LoRA.

fuse train --config train_extraction.yaml

Options

Option Description
--config PATH Required. YAML training config file

See Training Configuration for config file format.

fuse quantize

Convert a trained model to GGUF format.

fuse quantize --model ./output --output model.gguf --method q4_0

Options

Option Description
--model PATH Required. Path to the trained model directory
--output PATH Required. Output GGUF file path
--method TEXT Quantization method (default: q4_0)

Config file examples

Extraction config

model:
  model_name: "bartowski/Llama-3.2-1B-Instruct-GGUF"
  n_ctx: 2048
  temperature: 0.0

fields:
  name: str
  age: int
  job_title: str
  company: str

prompt_format: llama
max_tokens: 256

Training config

model_name: "unsloth/Llama-3.2-1B-Instruct"
output_dir: "./output/llama-extraction"
dataset_path: "./data/extraction_dataset.jsonl"

lora:
  r: 16
  alpha: 32
  dropout: 0.05

training:
  epochs: 3
  batch_size: 4
  learning_rate: 2.0e-4
  max_seq_length: 2048