docs(log): frank-large cutover verified live (MTP drafting, tools, e2e 32/32)
This commit is contained in:
@@ -93,6 +93,17 @@ timestamp: 2026-09-14
|
||||
cache, health 200) before the redeploy. The redundant 87 GiB assembled
|
||||
file is staged at `~/.AGENTTRASH/20260921-frank-large-assembled/`
|
||||
(dankclaw) awaiting purge approval.
|
||||
- **Verified after cutover (live).** Unit active; the journal shows the
|
||||
draft model loading and `model loaded` in ~37 s (warm page cache), then
|
||||
`listening on http://0.0.0.0:7770`. The MTP head is genuinely attached
|
||||
and drafting — `draft acceptance = 0.91667 (22/24), mean len = 2.83`
|
||||
and `0.83333 (30/36), mean len = 2.67` (n_max 2, i.e. both drafts
|
||||
accepted most of the time), not the silent main-model-speed mode.
|
||||
`/v1/models` → `frank-large`; a chat completion returns `content` plus
|
||||
extracted `reasoning_content`; a tool request returns a clean
|
||||
`tool_calls` array (`get_weather {"city":"Seattle"}`). The e2e gate
|
||||
failed once on the known-flaky `12-crawl` (an `app.plex.tv` link
|
||||
timed out — unrelated to this endpoint) and passed 32/32 on re-run.
|
||||
|
||||
## 2026-09-21 — node-dankclaw Prometheus target fixed (netavark host-bind quirk)
|
||||
|
||||
|
||||
Reference in New Issue
Block a user