From ccd87508995bf12835e7aee2ea4009b141ab96c6 Mon Sep 17 00:00:00 2001
From: Harald Hoyer <harald@hoyer.xyz>
Date: Thu, 21 May 2026 23:15:09 +0200
Subject: [PATCH] chore(halo): set spec-draft-p-min for coder model

Add a 0.74 confidence threshold so speculative drafting stops early
once the draft model's predicted token probability drops below it,
favoring shorter, higher-acceptance draft sequences.
---
 systems/x86_64-linux/halo/models.ini | 1 +
 1 file changed, 1 insertion(+)

diff --git a/systems/x86_64-linux/halo/models.ini b/systems/x86_64-linux/halo/models.ini
index 5b1bbac..bfc3fc9 100644
--- a/systems/x86_64-linux/halo/models.ini
+++ b/systems/x86_64-linux/halo/models.ini
@@ -18,6 +18,7 @@ c                   = 131072
 hf                  = unsloth/Qwen3.6-27B-MTP-GGUF:UD-Q8_K_XL
 spec-type           = ngram-simple,draft-mtp
 spec-draft-n-max    = 5
+spec-draft-p-min    = 0.74
 threads-batch       = 16
 temp                = 0.6
 top-p               = 0.95