Granite Guardian 4.1 8B · file← Granite Guardian 4.1 8B
BF16 safetensors
IBM's full-size original. Never run; we made the MLX copies from it. IBM, HF cache.
- Size
- 16.8 GB
- Status
- On disk
- Scored runs
- 0
ResultsNever scored
This file never completed a scored run. Its record below says why.
The recordHow it was made · how it was served · what went wrong · what happened to it
Rebuilt from the lab's records by Claude, an AI model. The results above come straight from the automated runs. How these pages are made.
| File | model-0000{1..4}-of-00004.safetensors (HF cache) |
| Format | BF16 safetensors, 16,761,144,464 B (4 shards) |
| Where from | downloaded 2026-09-09 ~10:37–10:45Z by snapshot_download inside the mlx_lm conversion |
| Status | on disk in the HF cache |
| Hugging Face | https://huggingface.co/ibm-granite/granite-guardian-4.1-8b |
Where it came from
- Repo:
ibm-granite/granite-guardian-4.1-8b, snapshotab01ccca5dcfb80246369a086a4a87a29198f5af(refs/main). - Files: four shards = 4,995,591,048 + 4,991,447,808 + 4,970,460,032 + 1,803,645,576 = 16,761,144,464 B (stat, 2026-10-23), plus config, tokenizer,
chat_template.jinja,model.sig, README. sha256 of the shards: the blob names are the LFS sha256s (e.g. shard 13e86d760…6913). - How: the first
mlx_lm convert --hf-path ibm-granite/granite-guardian-4.1-8bdownloaded a partial snapshot and failed withIncompleteSnapshotErrorafter quantising. The retry ransnapshot_download(SRC, max_workers=8)first. Blob mtimes 2026-09-09 11:37–11:49 local. - Date / who asked: 2026-09-09; Alan, "can we convert granite?".
How it was served
Never served. No run-log label.
| label | dates (UTC) | binary | flags | profile | why this label exists |
|---|---|---|---|---|---|
| — | — | — | — | — | source file only |
Decisions
- Downloaded as conversion input only. Kept because nothing has swept the HF cache's Guardian entry (it survived the 2026-09-16 and 2026-10-11 sweeps).
- Not a publish candidate (IBM's own upload).
What went wrong
- Partial-snapshot conversion on the first attempt: mlx_lm quantised an incomplete snapshot ("Quantized model with 3.906 bits per weight") and then failed to write.
What happened to it
Downloaded 2026-09-09 → used for mixed_3_6, 8-bit and bf16 MLX builds (2026-09-09) and the 4-bit build (2026-09-11) → on disk at .cache/huggingface/hub/models--ibm-granite--granite-guardian-4.1-8b (16 GB by du, checked 2026-10-23). Re-downloadable from HF at that revision.