Great model with real human needs
I tested the model, and to my surprise, it outperforms many of the so-called benchmark-setting models. Once again, Mistral proves itself as a leader. It would be great to see more technical details on research and the training process hope you guys make it public soon.
Same, did the same tests I run for all the models I look at. writing summaries, information extraction, menu navigation, integration of information from RAG into responses, basic logic and reasoning, EN/FR translation on the fly. And it performed at least as well as a Qwen 2.5 32B, if not a bit better here and there, and without the risk of getting random Chinese words mid-sentence.
Good job, Mistral!
Agreed. I switched from Qwen2.5-32b and used this instead and it's much better at staying on track
Excellent model so far. Definitelly feels like it punches way above its 24B, feels way more like at least 32B if not even bigger. Yet another great piece of tech from the OG lab which gave us Mixtral back in the day. Thank you!