Mitar

> You can speed up your CI a bit by having Percy download the browser on install and then caching dependencies. Or if you already have a Chromium browser available...

> There already exists plenty of docker images with browsers included and since Percy is typically a project dependency, there is little value in maintaining our own docker image for...

I made a [docker image](https://gitlab.com/mitar/mitarpage/-/blob/main/percy.dockerfile) (`registry.gitlab.com/mitar/mitarpage/percy/hash:27c5359f564bf64e6bed2973e348bc99c9e8176cb0f3fedb9ab600d9ea5dccdc`).

I use Chromium from a Docker image but the whole run still takes 23 minutes now to make snapshots of all 1000+ pages. Running that in parallel would save time.

JSON format in Ollama is converted to a JSON grammar and passed to llama.cpp. llama.cpp then restricts generated tokens to match the grammar. But there are some known performance issues...

Once those PRs are included in llama.cpp used by ollama, using grammar should be fast: https://github.com/ggerganov/llama.cpp/pull/6811 https://github.com/ggerganov/llama.cpp/pull/7424

After more testing it seems that only "format json" is slow while if I convert a custom JSON schema to grammar (see #5348), there is no slowdown. I think the...

Thanks for quick response. Is the background then a bug or a feature on canvg? Do you think it should be reported upstream?

I think it might be this problem: https://github.com/canvg/canvg/issues/314

Resizing does not yet work perfectly. The text gets moved. For example, see above example with width/height set to 42, and look where text is, and then try with 50....