Drifter4242
Drifter4242
Slicing this geometry only results in half of it being sliced. The geometry is somewhat complex, so I suspect something overflowed. I've attached the stl and screengrab of the result....
## Motivation KVCache wasn't working on MLX when using Kimi K2 causing slowness. This fixes the KVCache and uses it to store the last prompt generation (only). There are various...
Kimi-K2 Thinking uses tiktoken.model for its tokenizer, which wasn't being downloaded. This adds it to the default_patterns alongside tokenizer.model. I'm a bit confused why this isn't a problem for other...
When a client disconnects (e.g., user clicks stop in SillyTavern), the generation now stops instead of continuing to run. Flow: - API catches CancelledError and sends TaskCancelled command - Master...
## Motivation Saves the last launch settings, so that the next time you run exo it will default to the same launch settings. This is just a small quality of...
## Motivation I wanted to specify the temperature from the api. I added temperature along with top_p, logit bias, and frequency_penalty ## Changes Modified mlx_generate() to create samplers and logits...