liveBook · Manning

Levanter

As explained in Deep Learning with JAX

A framework for training large language models and other foundation models in JAX with emphasis on legibility, scalability, and reproducibility.

Need to train a large language model without giving up readable configuration? Levanter is a JAX framework for training LLMs and other foundation models, emphasizing legibility, scalability, and reproducibility. section 12.1.2

Levanter fits into the JAX ecosystem by combining:

  • Configurable model changes through edited parameters, rather than requiring direct source-code forks. section 12.1.2
  • Haliax, its companion named-tensor library, as an alternative to JAX’s xmap() mechanism. section 12.1.2
  • Large-scale execution across GPUs and TPUs, with demonstrated scale up to 20B parameters and a TPU v3-256 pod slice. section 12.1.2
  • A research-friendly position alongside alternatives such as MaxText, T5X, Jaxformer, JaxSeq, and Alpa. section 12.1.2

How Deep Learning with JAX covers it

How deep it goes
Characterizes goals, configuration, companion libraries, hardware, and demonstrated scale without detailing training procedures or model internals.
Examples
Uses Haliax, 20B-parameter models, and a TPU v3-256 pod slice as concrete examples.
What you can do after
Evaluate Levanter as a legible, scalable, reproducible JAX option for configurable foundation-model training.
How it connects
Connects JAX, Haliax, xmap, GPUs, TPUs, model parallelism, and configurable foundation-model training.
Cover of Deep Learning with JAX
liveBookmanning.com
Read Deep Learning with JAX on liveBook Get the book at manning.com

This explainer was generated by Manning's knowledge graph from the text of Deep Learning with JAX. The section links point to the parts of the book it draws on.