add gemma4 as a model for google - #225
Merged
Merged
Conversation
Signed-off-by: Manny Delgado <manny@duolingo.com>
mannydelgado
force-pushed
the
md/gemma-4-integration
branch
from
August 4, 2026 02:11
3d47191 to
6777b23
Compare
This was referenced Aug 10, 2026
Merged
handstandsam
added a commit
that referenced
this pull request
Aug 10, 2026
#225 added one Gemma 4 model to the google provider. Google serves three on the Gemini API free tier (12B dense, 26B-A4B MoE, 31B dense) and Ollama publishes the same three locally, so all six are now built in — pick one by id, no specs to look up. Two details worth calling out: - `gemma-4-31b-it` and `gemma-4-26b-a4b-it` are both real, distinct models. The mismatch between #225's yaml and its LLM_MODELS.md row read like a typo; it wasn't, and regenerating the doc silently dropped one of them. Both are here now. - max_output_tokens is 32768, not the 65536 the Gemini models use. Google publishes no output cap for Gemma 4; 32768 is the largest figure any host documents. The e2b/e4b edge builds are deliberately omitted from Ollama — 128K context and much weaker at the screenshot-driven tool calling Trailblaze's agent loop does. Note `gemma4:latest` resolves to e4b, so name a size explicitly. LLM_MODELS.md is generated from the provider yamls (`:docs:generator:run`).
handstandsam
added a commit
that referenced
this pull request
Aug 10, 2026
…a release (#229) * Add every hosted Gemma 4 variant to the built-in model list #225 added one Gemma 4 model to the google provider. Google serves three on the Gemini API free tier (12B dense, 26B-A4B MoE, 31B dense) and Ollama publishes the same three locally, so all six are now built in — pick one by id, no specs to look up. Two details worth calling out: - `gemma-4-31b-it` and `gemma-4-26b-a4b-it` are both real, distinct models. The mismatch between #225's yaml and its LLM_MODELS.md row read like a typo; it wasn't, and regenerating the doc silently dropped one of them. Both are here now. - max_output_tokens is 32768, not the 65536 the Gemini models use. Google publishes no output cap for Gemma 4; 32768 is the largest figure any host documents. The e2b/e4b edge builds are deliberately omitted from Ollama — 128K context and much weaker at the screenshot-driven tool calling Trailblaze's agent loop does. Note `gemma4:latest` resolves to e4b, so name a size explicitly. LLM_MODELS.md is generated from the provider yamls (`:docs:generator:run`). * Document how to use a model Trailblaze doesn't ship with Chasing model releases through Trailblaze releases is the wrong loop, and nothing required it — a workspace trailblaze.yaml can declare any model its provider serves, and it takes effect on the next run. That was documented only as a passing sentence inside the LLM reference. New 'Adding a Model' page covers it directly: how a workspace entry merges with the built-in list (by id — new ids add, known ids override field-by-field), the 131K/8K fallbacks you get by omitting the limits, recipes for hosted, local Ollama, and gateway models, and how to contribute a model back to the built-in registry (including regenerating LLM_MODELS.md, which is generated). Also fills a real gap: the page titled 'Configuration' documented Android instrumentation args and scripting timeouts and never mentioned trailblaze.yaml itself. It now opens with the file — where it lives, that every section is optional, what each top-level key does, the inline-or-ref shape, and one precedence table — with the existing sections kept below. * Point the External Config page at the trailblaze.yaml documentation Landing on the generated External Config page gave no way to find out what trailblaze.yaml itself is — the page covers the trails/config/ directory (targets, toolsets, tools) and only mentions the file in passing. It now opens with a pointer to the Configuration page, which documents the file's keys, workspace defaults, and precedence. Generator change (ExternalConfigDocsGenerator) + regenerated output.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Integrate Gemma4 as a model for google provider.
This is limiting in terms of context size but it does allow for some free usage which is nice for testing.