AI Radar

SmolLM

Available · Language · Milestone

SmolLM is a family of small open-weight language models from Hugging Face in 135M, 360M and 1.7B sizes, each released as a base and an instruction-tuned checkpoint under Apache 2.0. Hugging Face trained them on its SmolLM-Corpus and presents them as small enough to run locally on devices from smartphones to laptops. [1] [2] Primary source

Timeline of SmolLM →

Claims and evidence

  • Released Primary source[1]page date
  • SmolLM-Instruct v0.2 released Primary source[3]README, News: August 18, 2024 [4]Changelog, v0.2
  • Status Available Primary source[2]model repository (weights hosted on the Hub)
      • Input text Primary source[2]Model Summary; Limitations
      • Output text Primary source[2]Model Summary; Limitations
      • Open weights Yes Primary source[2]License: Apache 2.0
      • Context window 2048 tokens tokens Primary source[1]Training: Hyperparameters choice
      • Access Open-weights download, On device Primary source[1]How to run locally

      Lineage

      Predecessors

      No known predecessor.

      Successors

      Based on

      Not derived from another model.

      Variants and derived

      None recorded.

      Siblings

      None recorded.

      All ancestors

      None.

      All descendants

      Variants

      SmolLM-135M

      Same dates as SmolLM.

      • Variant Inline variant in this record.Trained on 600B tokens; architecture modelled on MobileLLM with grouped-query attention, per the blog. Primary source[1]TL;DR; Training

      Also known as: SmolLM-135M-Instruct

      Differs in:

      • Parameters: 135M Evidence not assessed [1]TL;DR

      SmolLM-360M

      Same dates as SmolLM.

      • Variant Inline variant in this record.Trained on 600B tokens; architecture modelled on MobileLLM with grouped-query attention, per the blog. Primary source[1]TL;DR; Training

      Also known as: SmolLM-360M-Instruct

      Differs in:

      • Parameters: 360M Evidence not assessed [1]TL;DR

      SmolLM-1.7B

      Same dates as SmolLM.

      • Variant Inline variant in this record.Trained on 1T tokens with a more conventional architecture than the two smaller sizes, per the blog. Primary source[1]TL;DR; Training [2]Training

      Also known as: SmolLM-1.7B-Instruct

      Differs in:

      • Parameters: 1.7B Evidence not assessed [1]TL;DR [2]Model Summary

      Related AI Radar coverage

      AI Radar coverage starts in June 2026; no coverage linked yet.

      All model releases from Hugging Face on AI Radar →