felafax-cli 0.1.16

CLI tool for interacting with the Felafax platform; fine-tuning and inference on non-NVIDIA chipsets.

Felafax CLI

A command-line interface for fine-tuning and managing large language models.

Installation

pip install felafax-cli

Quick Start

1. Login with your authentication token:

felafax-cli auth login --token <your-token>

2. Initialize a fine-tuning configuration:

felafax-cli tune init-config

3. Upload your training data:

felafax-cli files upload path/to/data.jsonl

4. Start fine-tuning:

felafax-cli tune start --model <model_name> --config config.yml --dataset <dataset_id>

Available Commands

Authentication

# Login with your token
felafax-cli auth login --token <your-token> [--force]

Model Management

# List available fine-tuned models
felafax-cli model list

# Start an interactive chat session with a model
felafax-cli model chat <model_id> [--system-prompt "Your custom prompt"]

# Get detailed information about a model
felafax-cli model info <model_id>

# Delete a fine-tuned model
felafax-cli model delete <model_id>

Training Management

# Initialize a new config file
felafax-cli tune init-config

# Start a new fine-tuning job
felafax-cli tune start \
    --model <model_name> \
    --config path/to/config.yml \
    --dataset <dataset_id>

# List all training jobs
felafax-cli tune list

# Check job status
felafax-cli tune status --job-id <id>

# Stop a running job
felafax-cli tune stop --job-id <id>

File Management

# List files in storage
felafax-cli files list [--prefix <prefix>] [--limit <number>]

# Upload training data
felafax-cli files upload <file_path>

# Delete a file
felafax-cli files delete <file_path>

Configuration Example

When you run felafax-cli tune init-config, it creates a YAML file with the following structure:

hyperparameters:
  learning_rate: 1.0e-05
  batch_size: 32
  n_epochs: 4
  warmup_ratio: 0.0
lora:
  enabled: false
  r: 8
  alpha: 8
  dropout: 0.0

Support

For more information, visit our documentation at https://docs.felafax.ai

License

MIT License