Thanks for the report that "mistral-alpha-large / Tekken / Dojo tutorial" output is the model dumping its training data instead of replying, which means our chat-template formatting is failing on your end. It's a known bug we just fixed; the next build will have it patched. Two quick questions while we sort it out:
1. Which LLM tier did you install (Lite / Standard / High / Ultra / Premium / Quality)?
2. In the in-game LLM tab, does the tier you picked say it's fully downloaded, and does the file size in LocalModels/LLM/ match? (e.g. Standard should be 7.5 GB a partial download can also cause garbage output.)