Qwen3-8B · file← Qwen3-8B
Q8_0
A larger, more precise copy for the August speed study. Never scored; deleted on 7 September as a duplicate. GGUF.
- Size
- 8.7 GB
- Status
- Deleted 7 Sep 2026
- Scored runs
- 0
This file never completed a scored run. Its record below says why.
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 | models/qwen3-8b-gguf/Qwen3-8B-Q8_0.gguf |
| Format | GGUF Q8_0, ~8.7 GB (8.71 GB / 8.11 GiB; exact bytes not on record) |
| Where from | downloaded before 2026-08-12 |
| Status | deleted 2026-09-07 17:36Z |
| Hugging Face | same repo, inferred |
Where it came from
- Size: 8.71 GB on the 2026-08-23 listing, 8.11 GiB in the 2026-08-12 lab note and the 2026-08-30 listing. llama-bench
model_size8,703,561,728 B (tensor data;eval/results_fleet_bench.json). Exact file bytes and sha256 not on record. - Date: present by 2026-08-12 (the lab note lists it); download command not on record.
- Why: the high-precision GGUF for the runtime study; the Aug 12 note reports it beside the Q4_K_M (pp512 438.0, tg128 37.2).
How it was served
| label | dates (UTC) | binary and version | ctx / -np / KV / -fa / --jinja / --reasoning-format | harness profile | why this label exists |
|---|---|---|---|---|---|
| none — never run on a suite | llama-bench only, 2026-08-23 13:06:56Z | llama-bench b10369 (6e62ba538) | -ngl 99, pp512 / tg128 defaults | n/a | fleet-wide speed sweep of 22 GGUFs |
It was never served by a llama-server under a harness label. The 2026-08-23 planning table that labels a row "qwen3-8b Q8" (22.70 GB, 4.33 s/turn) is probably the Q4_K_M router figure (see the Q4_K_M doc); which file was loaded is not on record.
Decisions
- Kept for the runtime study, then judged a duplicate of the Q4_K_M on disk-pressure grounds: the assistant's triage table lists
qwen3-8b-ggufas holding Q8_0 (8.1G) and Q4_K_M (4.7G), reclaimable about 5-8 GB. - Deleted 2026-09-07 17:36:31Z on Alan's "Go for it and what’s the blocker on Xcode? I can be installed from cli":
rm -f.../Qwen3-8B-Q8_0.gguf && echo " rm qwen3-8b Q8_0 (kept Q4_K_M)". - This was the first sweep of the programme; the RESTORE-script pattern began on 2026-09-08, so there is no RESTORE entry.
What went wrong
- None specific. Note the Aug 12 lab note quotes this file's speed as 37.2 t/s and the remeasurement on 2026-08-23 as 36.99, consistent.
- Standing checks: no run-log entries, so no partial-run, build-stamp or flag checks apply.
What happened to it
Downloaded (before 2026-08-12) -> llama-bench 2026-08-12 and 2026-08-23 -> deleted 2026-09-07 17:36:31Z (Alan's "Go for it", no RESTORE entry) -> re-downloadable from the Qwen3-8B GGUF repo (unsloth/Qwen3-8B-GGUF, presumed).