From a71812ea12cdc9e49c68b7f0430b35316a070258 Mon Sep 17 00:00:00 2001
From: localai-org-maint-bot
 <306269227+localai-org-maint-bot@users.noreply.github.com>
Date: Tue, 13 Oct 2026 07:32:42 +0000
Subject: [PATCH] chore(gallery): add Qwen3.8 GSQ-RCO IQ3_S

Complete the existing GSQ-RCO choices with the 3.50-bit build and its
vision projector. Pin downloads and verify both SHA256 values against
Hugging Face LFS metadata and download headers.

Correct the older quality description and document variant selection.

Assisted-by: Codex:gpt-6

(cherry picked from commit 70a2bd2c21cd18e39e8bdf3a62b3f54c8fc12f00, mudler/LocalAI#12651)
Signed-off-by: Ettore Di Giacinto <mudler@localai.io>
---
 gallery/index.yaml | 46 ++++++++++++++++++++++++++++++++++++++++++++--
 1 file changed, 44 insertions(+), 2 deletions(-)

diff --git a/gallery/index.yaml b/gallery/index.yaml
index 7d64c676a..4aefa75b0 100644
--- a/gallery/index.yaml
+++ b/gallery/index.yaml
@@ -6476,6 +6476,7 @@
     - model: qwen3.8-27b-gsq-rco-iq2-xs
     - model: qwen3.8-27b-gsq-rco-iq2-s
     - model: qwen3.8-27b-gsq-rco-iq3-xxs
+    - model: qwen3.8-27b-gsq-rco-iq3-s
     - model: qwen3.8-27b-q8
   url: github:mudler/LocalAI/gallery/virtual.yaml@master
   urls:
@@ -7173,8 +7174,8 @@
     - https://huggingface.co/Qwen/Qwen3.8-27B
     - https://huggingface.co/ISTA-DASLab/Qwen3.8-27B-GSQ-RCO-GGUF
   description: |
-    Qwen3.8-27B with ISTA DASLab's recommended 3.00-bit GSQ-RCO IQ3_XXS mixed
-    quantization. The 10.1 GB model provides the highest quality in this set
+    Qwen3.8-27B with ISTA DASLab's 3.00-bit GSQ-RCO IQ3_XXS mixed
+    quantization. The 10.1 GB model offers a smaller alternative to IQ3_S
     and retains the shared BF16 vision projector.
   last_checked: "2026-08-30"
   overrides:
@@ -7207,6 +7208,47 @@
     - filename: llama-cpp/mmproj/qwen3.8-27b-gsq-rco/mmproj-Qwen3.8-27B-BF16.gguf
       uri: huggingface://ISTA-DASLab/Qwen3.8-27B-GSQ-RCO-GGUF/mmproj-Qwen3.8-27B-BF16.gguf
       sha256: 13cb7bebccbd04afc8f4090cb949ecf8937cdf7377c5799b1a0c594e7c0d3e16
+- !!merge <<: *qwen3-8-27b
+  name: qwen3.8-27b-gsq-rco-iq3-s
+  variants: []
+  urls:
+    - https://huggingface.co/Qwen/Qwen3.8-27B
+    - https://huggingface.co/ISTA-DASLab/Qwen3.8-27B-GSQ-RCO-GGUF
+  description: |
+    Qwen3.8-27B with ISTA DASLab's recommended 3.50-bit GSQ-RCO IQ3_S mixed
+    quantization. The 11.8 GB model retains the shared BF16 vision projector
+    for image input and uses the embedded chat template with llama.cpp.
+  last_checked: "2026-10-12"
+  overrides:
+    backend: llama-cpp
+    context_size: 262144
+    function:
+      automatic_tool_parsing_fallback: true
+      grammar:
+        disable: true
+    known_usecases:
+      - chat
+      - vision
+    mmproj: llama-cpp/mmproj/qwen3.8-27b-gsq-rco/mmproj-Qwen3.8-27B-BF16.gguf
+    options:
+      - use_jinja:true
+    parameters:
+      min_p: 0
+      model: llama-cpp/models/qwen3.8-27b-gsq-rco/Qwen3.8-27B-GSQ-RCO-IQ3_S.gguf
+      presence_penalty: 0
+      repeat_penalty: 1
+      temperature: 1
+      top_k: 20
+      top_p: 0.95
+    template:
+      use_tokenizer_template: true
+  files:
+    - filename: llama-cpp/models/qwen3.8-27b-gsq-rco/Qwen3.8-27B-GSQ-RCO-IQ3_S.gguf
+      uri: https://huggingface.co/ISTA-DASLab/Qwen3.8-27B-GSQ-RCO-GGUF/resolve/d562806dbafae37109975e970aae91b43e73b440/Qwen3.8-27B-GSQ-RCO-IQ3_S.gguf
+      sha256: 64b53b64c7aa39f20a7e54bd80582fe595b1d745624ee8a72e92508c0326d810
+    - filename: llama-cpp/mmproj/qwen3.8-27b-gsq-rco/mmproj-Qwen3.8-27B-BF16.gguf
+      uri: https://huggingface.co/ISTA-DASLab/Qwen3.8-27B-GSQ-RCO-GGUF/resolve/d562806dbafae37109975e970aae91b43e73b440/mmproj-Qwen3.8-27B-BF16.gguf
+      sha256: 13cb7bebccbd04afc8f4090cb949ecf8937cdf7377c5799b1a0c594e7c0d3e16
 - name: underdog-saluki-27b-1.0
   url: github:mudler/LocalAI/gallery/virtual.yaml@master
   urls: