What's Changed

• Expose idefics2 loader by @EricLBuehler in https://github.com/EricLBuehler/mistral.rs/pull/450
• Try auto dtypes based on compute cap by @EricLBuehler in https://github.com/EricLBuehler/mistral.rs/pull/453
• Fix dtype error for logit bias by @EricLBuehler in https://github.com/EricLBuehler/mistral.rs/pull/454
• Fix sequence prompt len for Phi3-V by @EricLBuehler in https://github.com/EricLBuehler/mistral.rs/pull/455
• Tune threshold for matmul via f16 by @EricLBuehler in https://github.com/EricLBuehler/mistral.rs/pull/457
• Improve short/long scaling precision for LongRope by @EricLBuehler in https://github.com/EricLBuehler/mistral.rs/pull/458
• Fix LongRope models position ids calculation by @EricLBuehler in https://github.com/EricLBuehler/mistral.rs/pull/459
• Update deps by @EricLBuehler in https://github.com/EricLBuehler/mistral.rs/pull/460
• Improve handling of errors in auto dtype selection by @EricLBuehler in https://github.com/EricLBuehler/mistral.rs/pull/461
• Add support for cross-gpu device mapping by @EricLBuehler in https://github.com/EricLBuehler/mistral.rs/pull/462
• Bump version to 0.1.21 by @EricLBuehler in https://github.com/EricLBuehler/mistral.rs/pull/463

Full Changelog: https://github.com/EricLBuehler/mistral.rs/compare/v0.1.20...v0.1.21