From 12cfe1e51905fa55dd473679b26794059202c581 Mon Sep 17 00:00:00 2001 From: "K. Isom" Date: Tue, 22 Sep 2026 11:20:15 -0700 Subject: [PATCH] dixie: size honcho-embed so long inputs fit one batch and one slot batch-size defaulted to 2048 (capping ubatch) and ctx 8192 over 4 slots left 2048 per slot, so Honcho inputs over 2048 tokens failed with 'input is too large to process'. Now batch/ubatch 8192, ctx 16384 over 2 slots = 8192/slot, matching Honcho's EMBEDDING.MAX_INPUT_TOKENS. Verified a 6191-token input. Co-Authored-By: Claude Opus 5.5 (1M context) --- hw/dixie/default.nix | 9 +++++++-- 1 file changed, 7 insertions(+), 2 deletions(-) diff --git a/hw/dixie/default.nix b/hw/dixie/default.nix index 3ef26eb..2fbdb1e 100644 --- a/hw/dixie/default.nix +++ b/hw/dixie/default.nix @@ -51,9 +51,14 @@ let model = ${modelsDir}/qwen3-embedding-0.6b/Qwen3-Embedding-0.6B-Q8_0.gguf embedding = true pooling = cls + ; A whole input must fit in one physical batch AND one slot. batch-size defaults to 2048 + ; and silently caps ubatch, and ctx is split across slots, so both were 2048 in practice: + ; Honcho's longer messages (up to its 8192-token EMBEDDING.MAX_INPUT_TOKENS) failed with + ; "input is too large to process". Size all three to 8192 per slot. + batch-size = 8192 ubatch-size = 8192 - ctx-size = 8192 - parallel = 4 + ctx-size = 16384 + parallel = 2 ''; # Vision (auxiliary.vision for all agents): its own server on :11433, loaded on first