Add Qwen 3 14B (GGUF Q4_K_M) model support
- Adds Qwen 3 14B constants and metadata to models.rs - Exports GgufModelConfig::qwen3_14b() constructor - Updates SUPPORTED_MODELS and SUPPORTED_MODEL_INFO with 14B entry
keypair34 committed
Apr 26, 2026 at 07:39 UTC
69d0d0f61b81cbce149e22731acdf9c473651728
3 files changed
+32
-2
.github/workflows/ci.yml
-2
@@ -2,9 +2,7 @@ name: CI
2
3
on:
4
push:
5
- branches: [main, development]
5
pull_request:
7
- branches: [main, development]
6
workflow_dispatch:
7
8
concurrency:
src/inference/engine.rs
+15
@@ -2018,6 +2018,21 @@ impl GgufModelConfig {
2018
}
2019
}
2020
2021
+ /// Qwen 3 14B (GGUF Q4_K_M) — ~8.4 GB.
2022
+ ///
2023
+ /// Strong reasoning and tool-calling model with extended thinking.
2024
+ /// Best all-around model for macOS with 16+ GB RAM.
2025
+ pub fn qwen3_14b() -> Self {
2026
+ Self {
2027
+ model_id: super::models::BARTOWSKI_QWEN3_14B_GGUF.into(),
2028
+ files: vec![super::models::QWEN3_14B_GGUF_FILE.into()],
2029
+ tok_model_id: None,
2030
+ display_name: "Qwen 3 14B (Q4_K_M)".into(),
2031
+ approx_memory: "~8.4 GB".into(),
2032
+ chat_template: None,
2033
+ }
2034
+ }
2035
+
2036
/// DeepSeek Coder v1 6.7B Instruct (GGUF Q4_K_M) — ~3.8 GB.
2037
///
2038
/// Strong code generation model using the `llama` GGUF architecture.
src/inference/models.rs
+17
@@ -93,6 +93,14 @@ pub const QWEN3_4B_GGUF_FILE: &str = "Qwen_Qwen3-4B-Q4_K_M.gguf";
93
pub const BARTOWSKI_QWEN3_8B_GGUF: &str = "bartowski/Qwen_Qwen3-8B-GGUF";
94
pub const QWEN3_8B_GGUF_FILE: &str = "Qwen_Qwen3-8B-Q4_K_M.gguf";
95
96
+/// Pre-quantized Qwen 3 14B (GGUF Q4_K_M) — strong reasoning and tool-calling model (~8.4 GB).
97
+///
98
+/// Qwen 3 uses extended thinking mode (`<think>…</think>`) for improved reasoning.
99
+/// Best all-around model for macOS with 16+ GB RAM. Full tool calling support.
100
+pub const BARTOWSKI_QWEN3_14B_GGUF: &str = "bartowski/Qwen_Qwen3-14B-GGUF";
101
+/// The specific GGUF filename.
102
+pub const QWEN3_14B_GGUF_FILE: &str = "Qwen_Qwen3-14B-Q4_K_M.gguf";
103
+
104
/// Pre-quantized Qwen 3 1.7B (GGUF Q4_K_M) — lightweight tool-calling model (~1.3 GB).
105
///
106
/// Smallest Qwen 3 variant with tool calling support. Suitable for mobile devices
@@ -121,6 +129,7 @@ pub const SUPPORTED_MODELS: &[&str] = &[
129
BARTOWSKI_QWEN25_3B_INSTRUCT_GGUF,
130
BARTOWSKI_QWEN3_4B_GGUF,
131
BARTOWSKI_QWEN3_8B_GGUF,
132
+ BARTOWSKI_QWEN3_14B_GGUF,
133
BARTOWSKI_QWEN3_1_7B_GGUF,
134
BARTOWSKI_QWEN25_CODER_7B_INSTRUCT_GGUF,
135
THEBLOKE_DEEPSEEK_CODER_6_7B_INSTRUCT_GGUF,
@@ -189,6 +198,14 @@ pub const SUPPORTED_MODEL_INFO: &[SupportedModelInfo] = &[
198
Best balance of quality and memory for macOS with 24+ GB RAM.",
199
expected_size_bytes: 5_131_567_104,
200
},
201
+ SupportedModelInfo {
202
+ id: BARTOWSKI_QWEN3_14B_GGUF,
203
+ name: "Qwen 3 14B (GGUF)",
204
+ org: "Qwen / Alibaba",
205
+ description: "Strong reasoning and tool-calling model with extended thinking (~8.4 GB). \
206
+ Best all-around model for macOS with 16+ GB RAM.",
207
+ expected_size_bytes: 9_001_753_632,
208
+ },
209
SupportedModelInfo {
210
id: BARTOWSKI_QWEN3_1_7B_GGUF,
211
name: "Qwen 3 1.7B (GGUF)",