michael/AI-Profile-Router
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Document unified Medium slots and dashboard capacity

Browse Source
This commit is contained in:
Mikei386
2026-09-20 18:52:07 +02:00
parent 9af719674b
commit 53320fa6c3
9 changed files with 94 additions and 57 deletions
Download Patch File Download Diff File Expand all files Collapse all files
docs/PREFILL_BATCH_BENCHMARK_20260830.md
+6 -4
View File
@@ -23,10 +23,12 @@ The larger logical batches 3072 and 4096 did not improve Medium at ubatch 128. T
## Historischer Medium-Zwei-Slot-Test
This was an A/B candidate, not the current production configuration. Production
was returned to **one slot** because concurrent Hermes requests did not behave
reliably enough. The 85:15 GPU split and batch / ubatch 2048 / 128 remain in
production because they also work with the single-slot profile.
This began as an A/B candidate and was initially returned to **one slot**
because concurrent Hermes requests did not behave reliably enough. On
19 September 2026, Medium was enabled again with two unified-KV slots for a
controlled OpenClaw live trial. The figures below remain the historical
benchmark rather than a new performance claim. The current Medium ubatch is
512; the 85:15 GPU split is unchanged.
Identical fresh 100,297-token prompt with a deterministic 256-token completion: