Granite Guardian 4.1 8B · file ← Granite Guardian 4.1 8B

Q4_0

IBM's own Q4_0, not the one we made: 59 out of 60 the first time we tested it, on 26 October. IBM official GGUF.

Size: 4.8 GB
Status: On disk
Scored runs: 2
Smallest Mac: 16 GB estimate, 8K context

Results 2 scored · 95% intervals

Suite	Score	95% interval	Passed	Runtime · profile	Recorded
screening granite-guardian-4.1-8b-q4_0-ibm-np1-c64k-b11846	98.3%	91.1–99.7	Passed 59/60	llama.cpp extraction	2026-10-26
screening_native granite-guardian-4.1-8b-q4_0-ibm-np1-c64k-b11846	93.3%	84.1–97.4	Passed 56/60	llama.cpp extraction	2026-10-26

Runs marked unenforced were measured on a suite that requires constrained decoding, on a server shown not to apply the schema: mlx_lm accepts it and ignores it, and a few other servers and models break it. Those numbers do not measure what the suite is designed to measure. They are kept and labelled rather than deleted.

Setup How it was run

Of record

The measurement of record for this model.

The run’s own note: IBM's own Q4_0 (revision bc78f09, sha256 checked), measured for the first time. Not the copy requantised here from IBM's Q4_K_M that the screen seat serves (a different file of the same size). Single slot at 64K, f16 KV, b11846. A judge model: only screening/1 and screening-native/1 run. llama-bench (contended): pp512 382.5, tg128 52.8 tok/s.

Server: 1 slot, 64K of context per request, KV cache f16 (default), flash attention on, 1 suite at a time
Runtime: llama.cpp b11846 (1a24df163), plain upstream
Speed: not recorded
Peak memory: not recorded
Weights: 4.8 GB file · sha256 9eb58ed2189197e0…
Coherence check: not run (judge model: the gate asks for prose, which a judge does not write)
Left out by plan: tagger, tagger_nullable, needle, grounding, enrichment, epistemics, synthesis, worldview, nullschema, toolcall, toolcall2, toolscale, tools, contract, chainwalk, shapegen, sitebuild
Machine: Mac Studio (M1 Max, 64 GB)
Released: 26 October 2026, bundle 2026-10-26-70dce6d6

The record How 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	`granite-guardian-4.1-8b-Q4_0.gguf` (IBM's own)
Format	Q4_0 GGUF, 4,824,740,928 B, sha256 `9eb58ed2…`: the same size as the requant below, but a different file
Where from	downloaded 2026-10-26 from IBM's GGUF repo, revision `bc78f09`, size and sha256 checked against HF
Status	on disk
Hugging Face	https://huggingface.co/ibm-granite/granite-guardian-4.1-8b-GGUF

Where it came from

field	value
repo	`ibm-granite/granite-guardian-4.1-8b-GGUF` (no commits since 2026-05-26)
revision	`bc78f0995361543a70438fe44a60bb7613fed2f0`
file	`granite-guardian-4.1-8b-Q4_0.gguf`
size	4,824,740,928 B (matches the HF listing)
sha256	`9eb58ed2189197e047a6fd2d21fcadbd9c012762653b75a3b8a6c5b7f3096798` (matches the HF LFS record)
downloaded	2026-10-26 ~21:10, `huggingface_hub` from the cortext-benchmarks venv
local path	`models/granite-guardian-4.1-8b/ibm-bc78f09/granite-guardian-4.1-8b-Q4_0.gguf`
asked by	Alan, 2026-10-26: "granite important as IBM are a company I plan on approaching"

Not the seat's file. The SCREEN seat serves Granite Guardian 4.1 8B Q4_0 (requantised here from IBM's Q4_K_M): the same size, sha256 d9bdbcc5…, made here by requantising IBM's Q4_K_M. IBM's file is quantised once.

How it was served

Machine: Mac Studio (M1 Max, 64 GB).

label	date	binary	flags	profile	why
`granite-guardian-4.1-8b-q4_0-ibm-np1-c64k-b11846`	2026-10-26 21:45–21:48	`llama.cpp-b11846` (b11846, `1a24df163`)	`-np 1 -c 65536 -fa on -ngl 999 --cache-ram 0 --jinja`, f16 KV, on a bench port beside the fleet	extraction, `--no-coherence` (a judge)	OKF a lab note §8, pre-registered

Decisions

2026-10-26: downloaded and measured (plan §8). Whether the SCREEN seat moves to this file, with f16 KV, is Alan's call.

What went wrong

None.

What happened to it

Downloaded 2026-10-26 → measured 2026-10-26 → on disk. Re-downloadable from the repo and revision above.