• Joined on 2026-02-17
grenade commented on issue helexa/helexa#252 2026-08-14 15:26:15 +00:00
A cortex cold-load of the 27B produces an unservable model (catalogue has no quant)

Fixed and verified in production

quant = "q6k" added to the Qwen3.6-27B catalogue entry, and vram_mb corrected from 54000 (the bf16 checkpoint size) to 42000 (measured q6k residency across…

grenade pushed to fix/catalogue-quant-doc at helexa/helexa 2026-08-14 15:26:01 +00:00
4fc6e1d953 docs(catalogue): quant is required whenever a model is served quantised
grenade created branch fix/catalogue-quant-doc in helexa/helexa 2026-08-14 15:26:01 +00:00
grenade opened issue helexa/helexa#252 2026-08-14 15:17:06 +00:00
A cortex cold-load of the 27B produces an unservable model (catalogue has no quant)
grenade pushed to fix/drop-image-text-hint at helexa/helexa 2026-08-14 14:41:02 +00:00
b39e311d83 fix(web): drop the image lettering warning — it is wrong most of the time
grenade created branch fix/drop-image-text-hint in helexa/helexa 2026-08-14 14:41:02 +00:00
grenade pushed to main at grenade/c630 2026-08-14 13:54:45 +00:00
fc4d7d7ceb Stop kernel upgrades writing entries that cannot boot
grenade pushed to main at grenade/c630 2026-08-14 13:41:08 +00:00
36dc54e8db Keep the DSPs out of the initramfs, and let the headphone port be chosen
grenade pushed to main at grenade/c630 2026-08-14 13:17:57 +00:00
477d43aacd Survive kernel updates on a /boot this small
grenade commented on issue helexa/helexa#251 2026-08-13 10:52:29 +00:00
Implement Qwen3.8-27B support

What a single-GPU fallback would actually cost

Worth stating explicitly, because the acceptance criteria above understate it. ISQ and TP are not two independent conveniences — they are a…

grenade commented on issue helexa/helexa#251 2026-08-13 10:44:58 +00:00
Implement Qwen3.8-27B support

Correction: case B overstates the tensor-parallel cost

The qwen3_moe branch above reads as though TP would be off the table and calls building it "a real piece of work". Both overstate…

grenade commented on issue helexa/helexa#250 2026-08-13 10:40:17 +00:00
Serve Qwen3.8-27B

Implementation now tracked separately in #251, which carries the decision tree for what happens once model_type is known.

This issue stays as the preparation record — the tooling built, the…

grenade opened issue helexa/helexa#251 2026-08-13 10:40:06 +00:00
Implement Qwen3.8-27B support
grenade commented on issue helexa/helexa#250 2026-08-13 09:33:05 +00:00
Serve Qwen3.8-27B

Release slipped to 2026-08-14

The 2026-08-12 18:00 date passed without publication. Qwen/Qwen3.8-27B is now expected a day later.

The watcher has been restarted with a 7-day window (was…

grenade pushed to main at helexa/helexa 2026-08-12 14:19:39 +00:00
bdf102db01 Merge branch 'fix/await-model-placeholder'
a97eb7e5e0 fix(script): a reserved repo id is not a published model
Compare 2 commits »
grenade pushed to fix/await-model-placeholder at helexa/helexa 2026-08-12 13:55:27 +00:00
a97eb7e5e0 fix(script): a reserved repo id is not a published model
grenade created branch fix/await-model-placeholder in helexa/helexa 2026-08-12 13:55:27 +00:00
grenade pushed to main at helexa/helexa 2026-08-11 11:38:20 +00:00
586a99e871 Merge branch 'fix/nginx-absolute-redirect'
5e928b1de5 fix(nginx): relative redirects, so directory URLs are reachable
Compare 2 commits »
grenade pushed to fix/nginx-absolute-redirect at helexa/helexa 2026-08-11 10:58:29 +00:00
5e928b1de5 fix(nginx): relative redirects, so directory URLs are reachable
grenade created branch fix/nginx-absolute-redirect in helexa/helexa 2026-08-11 10:58:29 +00:00