[Promote] From Fireworks staging to cookbook

[bot-fireworks-ai]

Automated public repo promotion from fw-ai/fireworks.

Source staging path: public-repos/cookbook
Source staging SHA: fd1b56675625a72582e07471a45ae40563158fc4

[cursor]

Stale comment

Risk: medium. Not approved; human review is needed. The new reasoning_effort constructor path remaps training prefixes on shared GLM-5.2/5.3 renderers (including folding 5.2 low/medium onto High), so a human should confirm serving parity and that the default Max prefix is unchanged when the argument is omitted. Cursor Bugbot and Cursor Security Agent were not present after the first check poll.

  

_{Sent by Cursor Approval Agent: Pull Request Approver}

[cursor]

Stale comment

Risk: medium. Not approved; human review is needed. Shared GLM-5.2/5.3 renderers now remap reasoning_effort onto training prefixes (including folding 5.2 low/medium onto High), so a human should confirm serving parity and that omitting the argument still leaves the default Max prefix unchanged. Cursor Bugbot and Cursor Security Agent were not present after the first check poll.

  

_{Sent by Cursor Approval Agent: Pull Request Approver}

[cursor]

Stale comment

Risk: medium. Not approved; human review is needed. Shared GLM-5.2/5.3 renderers still remap reasoning_effort onto training prefixes (including folding 5.2 low/medium onto High), and Harbor recipes now expose epochs, off-policy headroom, and Tito artifact collection—confirm serving parity and that omitted-effort / recipe defaults stay unchanged. Cursor Bugbot and Cursor Security Agent were not present after the first check poll.

  

_{Sent by Cursor Approval Agent: Pull Request Approver}

[cursor]

Stale comment

Risk: medium. Not approved; human review is needed. Shared GLM-5.2/5.3 renderers still remap reasoning_effort onto training prefixes (including folding 5.2 low/medium onto High), and Harbor recipes now expose epochs, off-policy headroom, and Tito artifact collection—confirm serving parity and that omitted-effort / recipe defaults stay unchanged. Cursor Bugbot and Cursor Security Agent were not present after the first check poll.

  

_{Sent by Cursor Approval Agent: Pull Request Approver}

[cursor]

Risk: medium. Not approved; human review is needed. Shared Kimi K2.5/K2.6 renderers now resolve <|media_pad|> image placeholders (previously K2.7-only), and GLM-5.2/5.3 still remap reasoning_effort onto training prefixes—confirm tokenizer/unk handling and that omitted-effort plus Harbor recipe defaults stay unchanged. Cursor Bugbot and Cursor Security Agent were not present after the first check poll.

  

_{Sent by Cursor Approval Agent: Pull Request Approver}