kamma/llama-runner
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

added support for turboquant

Browse Source
This commit is contained in:
Radek Davidek 2026-03-29 17:45:06 +02:00
parent 71d370b63a
commit 1e6b33175a
3 changed files with 23 additions and 23 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
config.properties
View File

@ -1,5 +1,5 @@
#Llama Runner Configuration
#Sun Mar 29 16:13:49 CEST 2026
#Sun Mar 29 17:31:07 CEST 2026
windowHeight=1189
windowWidth=711
windowX=1849

40
profiles.json
View File

@ -6,8 +6,8 @@
"threads": 99,
"flashAttention": true,
"kvUnified": true,
"cacheTypeK": "q8_0",
"cacheTypeV": "q8_0",
"cacheTypeK": "turbo3",
"cacheTypeV": "turbo3",
"temperature": 0.6,
"topP": 0.95,
"topK": 20,
@ -27,8 +27,8 @@
"threads": 99,
"flashAttention": true,
"kvUnified": true,
"cacheTypeK": "q8_0",
"cacheTypeV": "q8_0",
"cacheTypeK": "turbo3",
"cacheTypeV": "turbo3",
"temperature": 0.6,
"topP": 0.95,
"topK": 20,
@ -48,8 +48,8 @@
"threads": 99,
"flashAttention": true,
"kvUnified": true,
"cacheTypeK": "q8_0",
"cacheTypeV": "q8_0",
"cacheTypeK": "turbo3",
"cacheTypeV": "turbo3",
"temperature": 0.6,
"topP": 0.95,
"topK": 20,
@ -69,8 +69,8 @@
"threads": 99,
"flashAttention": true,
"kvUnified": true,
"cacheTypeK": "q8_0",
"cacheTypeV": "q8_0",
"cacheTypeK": "turbo3",
"cacheTypeV": "turbo3",
"temperature": 0.6,
"topP": 0.95,
"topK": 20,
@ -90,8 +90,8 @@
"threads": 99,
"flashAttention": true,
"kvUnified": true,
"cacheTypeK": "q8_0",
"cacheTypeV": "q8_0",
"cacheTypeK": "turbo3",
"cacheTypeV": "turbo3",
"temperature": 0.6,
"topP": 0.95,
"topK": 20,
@ -132,8 +132,8 @@
"threads": 99,
"flashAttention": true,
"kvUnified": true,
"cacheTypeK": "q8_0",
"cacheTypeV": "q8_0",
"cacheTypeK": "turbo3",
"cacheTypeV": "turbo3",
"temperature": 0.6,
"topP": 0.95,
"topK": 20,
@ -153,8 +153,8 @@
"threads": 99,
"flashAttention": true,
"kvUnified": true,
"cacheTypeK": "q8_0",
"cacheTypeV": "q8_0",
"cacheTypeK": "turbo3",
"cacheTypeV": "turbo3",
"temperature": 0.6,
"topP": 0.95,
"topK": 20,
@ -174,8 +174,8 @@
"threads": 99,
"flashAttention": true,
"kvUnified": true,
"cacheTypeK": "q8_0",
"cacheTypeV": "q8_0",
"cacheTypeK": "turbo3",
"cacheTypeV": "turbo3",
"temperature": 0.6,
"topP": 0.95,
"topK": 20,
@ -195,8 +195,8 @@
"threads": 99,
"flashAttention": true,
"kvUnified": true,
"cacheTypeK": "q8_0",
"cacheTypeV": "q8_0",
"cacheTypeK": "turbo3",
"cacheTypeV": "turbo3",
"temperature": 0.6,
"topP": 0.95,
"topK": 20,
@ -216,8 +216,8 @@
"threads": 99,
"flashAttention": true,
"kvUnified": true,
"cacheTypeK": "q8_0",
"cacheTypeV": "q8_0",
"cacheTypeK": "turbo3",
"cacheTypeV": "turbo3",
"temperature": 0.6,
"topP": 0.95,
"topK": 20,

4
src/main/java/cz/kamma/llamarunner/Main.java
View File

@ -773,7 +773,7 @@ public class Main extends JFrame {
gbc.gridx = 1;
gbc.weightx = 1.0;
cacheTypeKComboBox = new JComboBox<>(new String[] { "bf16", "f16", "f32", "f8", "q8_0" });
cacheTypeKComboBox = new JComboBox<>(new String[] { "bf16", "f16", "f32", "f8", "q8_0", "turbo3", "turbo4" });
cacheTypeKComboBox.setSelectedIndex(0);
cacheTypeKComboBox.addActionListener(e -> updateCommandPreview());
panel.add(cacheTypeKComboBox, gbc);
@ -785,7 +785,7 @@ public class Main extends JFrame {
gbc.gridx = 1;
gbc.weightx = 1.0;
cacheTypeVComboBox = new JComboBox<>(new String[] { "bf16", "f16", "f32", "f8", "q8_0" });
cacheTypeVComboBox = new JComboBox<>(new String[] { "bf16", "f16", "f32", "f8", "q8_0", "turbo3", "turbo4" });
cacheTypeVComboBox.setSelectedIndex(0);
cacheTypeVComboBox.addActionListener(e -> updateCommandPreview());
panel.add(cacheTypeVComboBox, gbc);