From 819e49482ed49a4d2983611d71ac71e7ffbf3197 Mon Sep 17 00:00:00 2001
From: localai-org-maint-bot
 <306269227+localai-org-maint-bot@users.noreply.github.com>
Date: Fri, 23 Oct 2026 20:04:42 +0000
Subject: [PATCH] feat(gallery): add official Flint INT4 variant

Offer the publisher Q4_1 export alongside the existing Flint builds.
Pin the weights and document explicit variant installation.

Assisted-by: Codex:gpt-6
---
 docs/content/features/model-gallery.md | 14 ++++++++
 gallery/index.yaml                     | 42 ++++++++++++++++++++++
 gallery/models/index.yaml              | 50 ++++++++++++++++++++++++++
 3 files changed, 106 insertions(+)

diff --git a/docs/content/features/model-gallery.md b/docs/content/features/model-gallery.md
index 56dde2118..a8a6d5c9b 100644
--- a/docs/content/features/model-gallery.md
+++ b/docs/content/features/model-gallery.md
@@ -44,6 +44,20 @@ Both views store the view, search, filter, and selection in the URL. Installing
 from Explore does not move you away from the catalog; the entry updates in place
 when the operation finishes.
 
+### Flint INT4 GGUF
+
+[Recursive's official Flint Q4_1 build](https://huggingface.co/RecursiveSI/flint-int4-gguf)
+is available as a variant of `flint` for the llama.cpp backend.
+The 26 GB GGUF preserves the trained INT4 codes and includes the chat template.
+The publisher recommends at least 48 GB of unified memory on Apple Silicon.
+This entry uses a 32,768-token default context.
+
+Install this build explicitly:
+
+```bash
+local-ai models install flint --variant flint-int4-q4-1
+```
+
 ### MiniCPM-V 4.7 1B
 
 [MiniCPM-V 4.7 1B](https://huggingface.co/openbmb/MiniCPM-V-4.7-1B)
diff --git a/gallery/index.yaml b/gallery/index.yaml
index 8491bd92b..7cf1b466b 100644
--- a/gallery/index.yaml
+++ b/gallery/index.yaml
@@ -77085,6 +77085,7 @@
   variants:
     - model: flint-q2
     - model: flint-q6
+    - model: flint-int4-q4-1
 - url: github:mudler/LocalAI/gallery/virtual.yaml@master
   name: flint-q2
   description: |
@@ -77161,3 +77162,44 @@
       top_p: 0.95
     template:
       use_tokenizer_template: true
+- url: github:mudler/LocalAI/gallery/virtual.yaml@master
+  name: flint-int4-q4-1
+  description: |
+    Flint is a coding-agent mixture-of-experts model from Recursive, with
+    39.8B total and about 2.5B active parameters per token. This official
+    Q4_1 GGUF preserves the trained INT4 codes from the MLX export, with
+    F32 floating tensors. The 26 GB file includes the tokenizer and chat
+    template. This entry uses a 32,768-token default context.
+  license: apache-2.0
+  urls:
+    - https://huggingface.co/RecursiveSI/flint
+    - https://huggingface.co/RecursiveSI/flint-int4-gguf
+  tags:
+    - llm
+    - gguf
+    - cpu
+    - gpu
+    - moe
+    - coding
+    - agent
+    - reasoning
+    - tools
+  files:
+    - filename: llama-cpp/models/RecursiveSI/flint-int4-gguf/flint-int4-q4_1.gguf
+      sha256: 43c07baccdba42eb7528a9630ee15699b1942c297291dc1b39a061f508ee1a4b
+      uri: https://huggingface.co/RecursiveSI/flint-int4-gguf/resolve/9c47151036bc2c34331142dda04488bceab2bc9f/flint-int4-q4_1.gguf
+  overrides:
+    backend: llama-cpp
+    context_size: 32768
+    known_usecases:
+      - chat
+    options:
+      - use_jinja:true
+    parameters:
+      min_p: 0
+      model: llama-cpp/models/RecursiveSI/flint-int4-gguf/flint-int4-q4_1.gguf
+      temperature: 1
+      top_k: 20
+      top_p: 0.95
+    template:
+      use_tokenizer_template: true
diff --git a/gallery/models/index.yaml b/gallery/models/index.yaml
index 4a448f245..a37d93a99 100644
--- a/gallery/models/index.yaml
+++ b/gallery/models/index.yaml
@@ -94790,6 +94790,7 @@
     variants:
       - model: flint-q2
       - model: flint-q6
+      - model: flint-int4-q4-1
 - kind: model
   id: flint-q2
   name: flint-q2
@@ -94882,3 +94883,52 @@
         top_p: 0.95
       template:
         use_tokenizer_template: true
+- kind: model
+  id: flint-int4-q4-1
+  name: flint-int4-q4-1
+  version: 0.1.0
+  description: |
+    Flint is a coding-agent mixture-of-experts model from Recursive, with
+    39.8B total and about 2.5B active parameters per token. This official
+    Q4_1 GGUF preserves the trained INT4 codes from the MLX export, with
+    F32 floating tensors. The 26 GB file includes the tokenizer and chat
+    template. This entry uses a 32,768-token default context.
+  license: apache-2.0
+  urls:
+    - https://huggingface.co/RecursiveSI/flint
+    - https://huggingface.co/RecursiveSI/flint-int4-gguf
+  tags:
+    - llm
+    - gguf
+    - cpu
+    - gpu
+    - moe
+    - coding
+    - agent
+    - reasoning
+    - tools
+  channel: gallery
+  policy:
+    requireSignature: false
+  schemaVersion: 1
+  model:
+    url: github:mudler/LocalAI/gallery/virtual.yaml@master
+    files:
+      - filename: llama-cpp/models/RecursiveSI/flint-int4-gguf/flint-int4-q4_1.gguf
+        sha256: 43c07baccdba42eb7528a9630ee15699b1942c297291dc1b39a061f508ee1a4b
+        uri: https://huggingface.co/RecursiveSI/flint-int4-gguf/resolve/9c47151036bc2c34331142dda04488bceab2bc9f/flint-int4-q4_1.gguf
+    overrides:
+      backend: llama-cpp
+      context_size: 32768
+      known_usecases:
+        - chat
+      options:
+        - use_jinja:true
+      parameters:
+        model: llama-cpp/models/RecursiveSI/flint-int4-gguf/flint-int4-q4_1.gguf
+        temperature: 1
+        top_k: 20
+        top_p: 0.95
+        min_p: 0
+      template:
+        use_tokenizer_template: true