Would be very curious how RL benchmarks shake out vs M5 Pro/Max.
Doubt local inference is the target use case near nearly as much as post-training. I could totally see something like this being super appealing for a startup looking to do some fine-tuning/distillation to tune a small open-weight model for a narrow use case.
Doubt local inference is the target use case near nearly as much as post-training. I could totally see something like this being super appealing for a startup looking to do some fine-tuning/distillation to tune a small open-weight model for a narrow use case.