Frank
a5fe045b58
Structural repair: - 13 pages were the same document twice (each exactly 2x its HEAD size), a second frontmatter block whose closing fence had been glued onto a body line by a frontmatter rewrite that re-read its own output. Un-doubled by cutting at the second frontmatter and keeping the copy whose body is a superset; the orphan census reports 0 lost tokens against HEAD. - log.md: one entry per date, bare `## YYYY-MM-DD` headings, newest first — 66 dated headings merged into 34, titles demoted to bold leads, every content line verified present. The file had been descending at the head and ascending at the tail. Ownership moves: - New `LLM/frank-medium.md` is the locator for the retired Dankserver endpoint: era windows quoted from each era page's own banner, no numbers restated, and the two unresolved date conflicts recorded instead of silently resolved. `frank-archon.md` links to it instead of restating the pre-rename window. - The four Pipeline-Parallelism stubs (frontmatter only, no body) are gone; `Multi-Node Runtimes.md` owns runtime construction, launch, mixed-vendor verdicts and cross-split spec decode, and the four links that pointed at the stubs were repointed at it or at the drafter page. - Authentik's three stub pages fold into `Authentik.md`, with the process detail and the outpost reference split out so the parent stays under the length hint. Verified corrections (host census + live probe, 2026-10-19): - The MXFP4 checkpoint two pages called the "on-disk rollback" is gone: /home/dank/models/ holds StillDeadcode, rad and tpl.jinja. Stated once, on the era page that owns that payload; the swap page now links to it. - frank-medium-radiance-mxfp4: the 4.24/7 accept rate belongs to the DFlash2 arm measured on the FP8 target, not to the MXFP4 era; the 3-family soak run census is kept once (155 requests) instead of twice (~120 and 155). - Fleet shapes re-confirmed against the live endpoints before touching any page: 3x200000 fp8 KV, 2x200704, 2x262144. Index rows are regenerated, not hand-carved. Net -1,849 vault lines.
57 lines
1.7 KiB
Markdown
57 lines
1.7 KiB
Markdown
---
|
|
type: reference
|
|
title: frank-dragoon ubatch sweep
|
|
description: "The wide ubatch sweep for a deployed ROCmFPX shape, run through the standardized harness."
|
|
tags:
|
|
- llama.cpp
|
|
- rocm
|
|
- benchmark
|
|
- ubatch
|
|
timestamp: 2026-09-01
|
|
---
|
|
|
|
# frank-dragoon ubatch sweep
|
|
|
|
This benchmark uses the active `frank-dragoon` model on Dankclaw with
|
|
`llama-bench`. It does not use `llama-server`, HTTP requests, or server-log
|
|
scraping. Stop the production service before running it and start the service
|
|
again afterward.
|
|
|
|
The old custom harnesses are retained in `.AGENTTRASH/bench-scripts-20260901/`
|
|
for auditability only. Their results are invalid and must not be used.
|
|
|
|
## Sweep
|
|
|
|
The requested points are `256, 512, 768, 1024, 1280, 1536, 1792, 2048,
|
|
3072, 4096` at 8K context, with `batch` equal to `ubatch` at every point.
|
|
Each point runs three 128-token repetitions by default.
|
|
|
|
Run on Dankclaw from the repository checkout:
|
|
|
|
```bash
|
|
sudo systemctl stop frank-dragoon
|
|
./scripts/bench-ubatch.sh
|
|
sudo systemctl start frank-dragoon
|
|
```
|
|
|
|
The script uses the active Q6 AGENT model, an 8K prompt workload, equal
|
|
`batch` and `ubatch` values, and `llama-bench` JSONL output. The sweep is:
|
|
```text
|
|
256, 512, 768, 1024, 1280, 1536, 1792, 2048, 3072, 4096
|
|
```
|
|
|
|
The GPU is shared with live workloads; do not run with the production service
|
|
up.
|
|
|
|
## Results
|
|
|
|
No valid results are recorded here. All results from the retired custom
|
|
harnesses are explicitly discarded. Record the raw `llama-bench` JSONL output
|
|
from the new harness here after a clean run.
|
|
|
|
## See also
|
|
|
|
- [LLM fleet index](index.md)
|
|
- [Deployment](deployment.md)
|
|
- [[frank-dragoon]] — the kingjones MTP era this sweep ran against is an Eras row there (its page was folded in 2026-10-18)
|