v0.5.2

What's Changed

A few fixes and new additions:

• Support for CohereAI's command-r model: Currently, GGUF is unsupported. You can load the base model with --load-in-4bit or --load-in-smooth if you have an RTX 20xx series (or sm_75).
• Fix an issue where some GPU blocks were missing. This should give a significant boost to how much context you can use.
• Fix logprobs when -inf with some models.

Full Changelog: v0.5.1...v0.5.2