e5d71f0631ee72015b7b1fbcce44ad421c83430a
- Author
- TheEdgeOfRage <git@theedgeofrage.com>
- Committer
- TheEdgeOfRage <git@theedgeofrage.com>
- Date
Message
Add Qwen3.6 model to llama-server config
Diff
1diff --git a/dot_config/llama-server/models.ini b/dot_config/llama-server/models.ini
2index 7ef037894dd9e30bf169c992261db908ff803ddf..586cca12f4be00a698b029ba849b55a6dde46bad 100644
3--- a/dot_config/llama-server/models.ini
4+++ b/dot_config/llama-server/models.ini
5@@ -1,4 +1,4 @@
6-[qwen35-4b]
7+[qwen3.5-4b]
8 hf = unsloth/Qwen3.5-4B-MTP-GGUF:UD-Q4_K_XL
9 n-gpu-layers = 99
10 ctx-size = 8192
11@@ -6,4 +6,43 @@ flash-attn = on
12 parallel = 1
13 spec-type = draft-mtp
14 spec-draft-n-max = 6
15-chat-template-kwargs = {"enable_thinking":true}
16+reasoning = on
17+
18+[qwen3.6-27b]
19+hf = unsloth/Qwen3.6-27B-MTP-GGUF:Q4_K_M
20+alias = qwen3.6-27b
21+n-gpu-layers = 99
22+flash-attn = on
23+jinja = true
24+ctx-size = 65536
25+temp = 0.9
26+top-p = 0.95
27+top-k = 20
28+min-p = 0.0
29+parallel = 1
30+spec-type = draft-mtp
31+spec-draft-n-max = 2
32+repeat-penalty = 1.0
33+reasoning = on
34+
35+# [gemma4-26b]
36+# hf = unsloth/gemma-4-26B-A4B-it-GGUF:UD-Q4_K_M
37+# hf-draft = unsloth/gemma-4-26B-A4B-it-GGUF:Q8_0-MTP
38+# alias = gemma4-26B-A4B
39+# n-gpu-layers = 99
40+# n-gpu-layers-draft = 99
41+# flash-attn = on
42+# ctx-size = 65536
43+# temp = 0.9
44+# top-p = 0.95
45+# top-k = 64
46+# spec-type = draft-mtp
47+# spec-draft-n-max = 2
48+# fit = off
49+# reasoning = on
50+#
51+# [lfm2-24b]
52+# hf = LiquidAI/LFM2-24B-A2B-GGUF:Q4_K_M
53+# n-gpu-layers = 99
54+# ctx-size = 16384
55+# flash-attn = on
56diff --git a/dot_local/bin/executable_pkgbuild-review b/dot_local/bin/executable_pkgbuild-review
57index 7ef057d3806d49a9b565fe398e75c33967693347..11f2366f483bb3d4d9b603f34c95f2c9cf4bab23 100755
58--- a/dot_local/bin/executable_pkgbuild-review
59+++ b/dot_local/bin/executable_pkgbuild-review
60@@ -2,7 +2,7 @@
61 set -euo pipefail
62
63 SERVER_URL="${PKGBUILD_REVIEW_URL:-http://127.0.0.1:8080}"
64-MODEL="${PKGBUILD_REVIEW_MODEL:-qwen35-4b}"
65+MODEL="${PKGBUILD_REVIEW_MODEL:-qwen3.5-4b}"
66 CACHE_DIR="${PKGBUILD_REVIEW_CACHE:-$HOME/.cache/aur-review}"
67 REAL_EDITOR="${REAL_EDITOR:-${EDITOR:-vi}}"
68