Gemma

Google’s open-weightⓘopen weightsweightsA model release that publishes the trained parameters under some downloadable license, distinct from "open source" which (per OSAID) also requires data and training-code openness. Chat about this → Open full entry → line, launched February 2024 as a 2B/7B model pair, since extended to 9B and 27B base sizes (Gemma 2) and to multimodal and code-specific variants. The architecture and tokenizerⓘtokenizerdataThe component that splits raw text into discrete units (tokens) the model can process, usually using a learned subword vocabulary like Byte-Pair Encoding. Chat about this → Open full entry → inherit from Gemini research; the goal of the release is to give researchers and developers access to Google’s open-weightⓘopen weightsweightsA model release that publishes the trained parameters under some downloadable license, distinct from "open source" which (per OSAID) also requires data and training-code openness. Chat about this → Open full entry → contribution.

The license terms are notable. Gemma is source-availableⓘsource-availableweightsA license category that lets users read and modify the code or weights but imposes restrictions (use limits, non-compete, MAU thresholds) that exclude it from the strict open-source definition. Chat about this → Open full entry →, not open source: an acceptable-useⓘacceptable-usegovernanceLicense or terms-of-service clauses that prohibit certain uses (weapons, surveillance, harassment, child sexual abuse material), common on open-weight licenses but rejected by the strict open-source definition. Chat about this → Open full entry → clause and a license-revocation hook distinguish it from Apache-licensed weights. The use restrictions parallel LlamaⓘLlamaweightsMeta's open-weight model family, the most widely deployed open release through 2024 to 2026, released under the source-available Community License with an MAU cap and acceptable-use clause. Chat about this → Open full entry →’s; the revocation hook is more aggressive.