Experimental model, using a limarp qlora trained at 10k ctx length (greater than size of the longest limarp sample when tokenized via mistral’s tokenizer) on mistralai/Mixtral-8x7B-v0.1 and then fused to mistralai/Mixtral-8x7B-Instruct-v0.1 at 0.5 weight. Note that all modules were trained, including ‘gate’. There are some reports that perhaps training the ‘gate’ module isn’t fully functional at the moment. In cursory testing this appears to obey the limarp alpaca prompt format correctly. Not extensively tested for quality, YMMV. Would try with temp ~1.5-2 and min-p of ~0.03-0.05 since mixtral does appear to be highly confident on its responses. The intended prompt format is the Alpaca instruction format of LimaRP v3: Due to the inclusion of LimaRP v3, it is possible to append a length modifier to the response instruction sequence, like this: This has an immediately noticeable effect on bot responses. The available lengths are: micro, tiny, short, medium, long, massive, huge, enormous, humongous, unlimited. The recommended starting length is medium. Keep in mind that the AI may ramble or impersonate the user with very long messages. The model will show biases similar to those…
Модальности:
Генерация текста
Области применения:
Диалог / чат Следование инструкциям
Задача: Генерация текста
Автор: DS-Archive
Теги: mixtral, conversational, en, text-generation-inference
Лайков: 3 | Загрузок: 22
Описание основано на материалах HuggingFace. Перевод выполнен автоматически.
