v0.65.2

Fixed

• The CLI's local (LMSupply) completion provider now formats prompts with the model's chat template and honours thinking mode
  and ThrowOnTruncation. It used a hand-written User:/Assistant: transcript; generation now goes through FluxImprover.LMSupply.
  Default temperature (0.7) and the 2048-token limit are unchanged.
• IMetadataEnricher.EnrichBatchAsync's documentation no longer promises a token-optimised single request. It answers
  cached requests from the cache and makes one EnrichAsync call per other request, in order — as the tutorial says.

Dependencies

• Re-pinned sibling package(s) FluxCurator 0.18.0 -> 0.18.1, LMSupply.Captioner 0.131.0 -> 0.132.0, LMSupply.Core 0.131.0 -> 0.132.0, LMSupply.Embedder 0.131.0 -> 0.132.0, LMSupply.Generator 0.131.0 -> 0.132.0, LMSupply.Ocr 0.131.0 -> 0.132.0, LMSupply.Transcriber 0.131.0 -> 0.132.0.
• Re-pinned sibling package(s) LMSupply.Captioner 0.132.0 -> 0.133.0, LMSupply.Core 0.132.0 -> 0.133.0, LMSupply.Embedder 0.132.0 -> 0.133.0, LMSupply.Generator 0.132.0 -> 0.133.0, LMSupply.Ocr 0.132.0 -> 0.133.0, LMSupply.Transcriber 0.132.0 -> 0.133.0.
• Re-pinned sibling package(s) LMSupply.Captioner 0.133.0 -> 0.134.0, LMSupply.Core 0.133.0 -> 0.134.0, LMSupply.Embedder 0.133.0 -> 0.134.0, LMSupply.Generator 0.133.0 -> 0.134.0, LMSupply.Ocr 0.133.0 -> 0.134.0, LMSupply.Transcriber 0.133.0 -> 0.134.0.
• Re-pinned sibling package(s) LMSupply.Captioner 0.134.0 -> 0.136.0, LMSupply.Core 0.134.0 -> 0.136.0, LMSupply.Embedder 0.134.0 -> 0.136.0, LMSupply.Generator 0.134.0 -> 0.136.0, LMSupply.Ocr 0.134.0 -> 0.136.0, LMSupply.Transcriber 0.134.0 -> 0.136.0.
• Re-pinned sibling package(s) FluxImprover 0.22.0 -> 0.23.0, FluxImprover.LMSupply 0.22.0 -> 0.23.0.