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

added support for q5_0 cache quant

Browse Source
This commit is contained in:
rdavidek 2026-05-16 19:44:23 +02:00
parent 0acebfd315
commit d0b693d1a9
2 changed files with 6 additions and 6 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

8
src/main/java/cz/kamma/llamarunner/Main.java
View File

@ -814,8 +814,8 @@ public class Main extends JFrame {
gbc.gridx = 1; gbc.gridx = 1;
gbc.weightx = 1.0; gbc.weightx = 1.0;
cacheTypeKComboBox = new JComboBox<>(new String[] { "bf16", "f16", "f32", "f8", "q8_0", "turbo3", "turbo4" }); cacheTypeKComboBox = new JComboBox<>(new String[] { "bf16", "f16", "f32", "f8", "q8_0", "q5_0", "turbo3", "turbo4" });
cacheTypeKComboBox.setSelectedIndex(0); cacheTypeKComboBox.setSelectedIndex(4);
increaseComboBoxHeight(cacheTypeKComboBox); increaseComboBoxHeight(cacheTypeKComboBox);
cacheTypeKComboBox.addActionListener(e -> updateCommandPreview()); cacheTypeKComboBox.addActionListener(e -> updateCommandPreview());
panel.add(cacheTypeKComboBox, gbc); panel.add(cacheTypeKComboBox, gbc);
@ -827,8 +827,8 @@ public class Main extends JFrame {
gbc.gridx = 1; gbc.gridx = 1;
gbc.weightx = 1.0; gbc.weightx = 1.0;
cacheTypeVComboBox = new JComboBox<>(new String[] { "bf16", "f16", "f32", "f8", "q8_0", "turbo3", "turbo4" }); cacheTypeVComboBox = new JComboBox<>(new String[] { "bf16", "f16", "f32", "f8", "q8_0", "q5_0", "turbo3", "turbo4" });
cacheTypeVComboBox.setSelectedIndex(0); cacheTypeVComboBox.setSelectedIndex(4);
increaseComboBoxHeight(cacheTypeVComboBox); increaseComboBoxHeight(cacheTypeVComboBox);
cacheTypeVComboBox.addActionListener(e -> updateCommandPreview()); cacheTypeVComboBox.addActionListener(e -> updateCommandPreview());
panel.add(cacheTypeVComboBox, gbc); panel.add(cacheTypeVComboBox, gbc);

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

@ -39,8 +39,8 @@ public class ModelConfig implements Serializable {
this.threads = 99; this.threads = 99;
this.flashAttention = true; this.flashAttention = true;
this.kvUnified = true; this.kvUnified = true;
this.cacheTypeK = "bf16"; this.cacheTypeK = "q8_0";
this.cacheTypeV = "bf16"; this.cacheTypeV = "q8_0";
this.temperature = 0.6; this.temperature = 0.6;
this.topP = 0.95; this.topP = 0.95;
this.topK = 20; this.topK = 20;