sync: engine 0.1.7 — hub cache fix, MLX import fix, default-q6 profile#24
Conversation
…-q6 profile, raw token decode API
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: a48581a3ba
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
| if std::fs::rename(&real, dst).is_ok() { | ||
| return Ok(()); |
There was a problem hiding this comment.
In install_file, the owned-cache path renames the real hf-hub blob into model.base but leaves the snapshot symlink in .src/hf dangling until a later caller invokes cleanup_staging. If the process is interrupted after this return and before cleanup (for example during catalog SHA verification or sidecar writing), retries see a broken staged snapshot for the same repo and can remain wedged until the user manually deletes the staging directory. Please remove the repo staging immediately after a successful move, or otherwise avoid returning with a dangling hf-hub cache entry.
Useful? React with 👍 / 👎.
1 participant
Sync with engine 0.1.7 (path-mapped from the internal
v0.1.7tag — exact tag-to-tag diff, no drift).basert pullno longer duplicates pulled models in the global HF cache (staged download + rename, sha-gated cleanup, resume preserved; opt-outBASERT_KEEP_HF_SOURCES=1). Catalog updates.default-q6target (--target base-q6).baseRT_decode_token_raw()— length-preserving raw-byte token decode for byte-level BPE tokens that can contain NUL. No struct-layout changes, so binding size guards are unaffected.Merge with (or after) the v0.1.7 engine release.