fairseq2.models

The models module provides pre-trained models and model architectures for various tasks including language modeling, machine translation, and speech recognition.

The models module includes:

• LLaMA models

• Qwen models

• Mistral models

• Wav2Vec2 models

• NLLB translation models

• Model loading and configuration utilities

• Model hub interface for advanced operations

• fairseq2.models.hub
  • Quick Start
  • Core Classes
  • Global Functions
  • Working with Model Families
  • Advanced Usage
  • Error Handling
  • See Also
• fairseq2.models.qwen
  • Quick Start
  • Available Models
  • Model Hub
  • Model Configuration
  • Model Factory
  • Tokenizer
  • Interoperability
  • Sharding
  • Constants
  • Complete Examples
  • See Also

Quick Start

Loading a Model

from fairseq2.models import load_model
from fairseq2.device import get_default_device

device = get_default_device()
model = load_model("qwen3_0.6b", device=device)

Listing Available Models

# List models from command line
python -m fairseq2.assets list --kind model

Getting Model Information

>>> from fairseq2.assets import get_asset_store
>>> asset_store = get_asset_store()
>>> card = asset_store.retrieve_card("qwen25_7b")
>>> print(f"Model family: {card.field('model_family').as_(str)}")
Model family: qwen
>>> print(f"Architecture: {card.field('model_arch').as_(str)}")
Architecture: qwen25_7b

Supported Model Families

Please check src/fairseq2/models/ directories for supported model families. You can also list available models from command line:

# List models from command line
$ python -m fairseq2.assets list --kind model