resolve_gguf_path() now glob-searches the entire model cache and returns the first lexicographically-sorted match. In the HuggingFace cache this can include multiple snapshot revisions, so this can select an older (or otherwise unintended) GGUF file when several revisions are present. Consider limiting the search to a specific snapshot (e.g., the commit referenced by refs/main or by an optional revision parameter) and/or prefer the most recently-modified snapshot before picking the first shard.
Originally posted by @Copilot in #9 (comment)
resolve_gguf_path() now glob-searches the entire model cache and returns the first lexicographically-sorted match. In the HuggingFace cache this can include multiple snapshot revisions, so this can select an older (or otherwise unintended) GGUF file when several revisions are present. Consider limiting the search to a specific snapshot (e.g., the commit referenced by refs/main or by an optional revision parameter) and/or prefer the most recently-modified snapshot before picking the first shard.
Originally posted by @Copilot in #9 (comment)