We scan this tool’s changelog a few times a day and keep the changes that would alter your agent’s next action: 68 since January 2026, each linked to its source.
hfFor a running job, use `hf jobs stats <job-id>` or `fetch_job_metrics(job_id=..., follow=False)` to read the current metrics sample and return without waiting. Add `-f/--follow` or `follow=True` to stream until the job completes.
hf jobs stats <job-id>When maintaining a Hugging Face Job workflow, use `--attempts` to retry failures, `hf jobs rerun` to create a fresh Job from a saved spec, and `hf jobs scheduled reschedule` to change a schedule. Use `hf jobs expose` to replace exposed ports on a running Job.
hf jobs run --attempts 3 --detach python:3.12 python train.pyBefore calling ZeroGPU Spaces, check the remaining quota and reset time with `hf spaces zero-gpu-quota`.
When configuring an `hf` webhook to watch bucket items, pass `bucket:<owner>/<bucket>` to `--watch`.
hf webhooks create --url <url> --watch bucket:<owner>/<bucket>When you need to remove one exact cached file while keeping the repo cache usable, pass its `hf://` file URI to `hf cache rm`; it removes that file from every cached revision and deletes its blob only when no other cached file references it. The path must be exact, and file targets cannot be mixed with repository or revision targets.
hf cache rm hf://models/<owner>/<repo>/<exact-file-path> --dry-runWhen a UV script needs a specific runtime or launch configuration, declare optional settings in its PEP 723 `[tool.hf-jobs]` table and run `hf jobs uv run <script>`. Script values act as defaults; CLI flags override them, while `env`, `secrets`, `labels`, and `volumes` merge entry by entry.
hf jobs uv run <script.py>When passing secrets to a sandbox, use the name-only form to resolve values from the caller’s environment; however, v1.32.0 docs do not show a secrets option for `hf sandbox exec`. They document `--secrets` when creating a dedicated sandbox, then running commands with `hf sandbox exec`.
hf sandbox create -s <SECRET_NAME>When listing scheduled jobs, filter by status, labels, or name with `hf jobs scheduled ls`; migrate deprecated `-f`/`--filter` expressions to the corresponding flags because they are ignored and warn. In Python, pass label matches to `list_scheduled_jobs(labels=...)`.
hf jobs scheduled ls --status suspendedFor an existing bucket, set `private=True` to make it private or `private=False` to make it public. Use `update_bucket_settings()` or `hf buckets settings` with the matching visibility option.
hf buckets settings <owner>/<bucket> --privateWhen deploying a vLLM or SGLang endpoint on a multi-accelerator instance, set parallelism to use the allocated accelerators. Use `hf endpoints deploy` to configure its engine image, or `hf endpoints update` to retune a running endpoint.
hf endpoints deploy gpt-oss-120b-vllm --repo openai/gpt-oss-120b --framework custom --accelerator gpu --instance-size x8 --instance-type nvidia-h200 --region us-east-1 --vendor aws --engine vllm --custom-image vllm/vllm-openai:v0.23.0 --tensor-parallel-size 8