public_git
/
whisper.cpp
mirror of https://github.com/ggml-org/whisper.cpp.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

bindings-java : disable flash attention by default (#3445) 

This commit disables flash-attention for the Java binding test so that
the testFullTranscribe test passes.

Without this change the test was failing because the expected output
mismatches after the flash-attention change:
```console
<And so my fellow Americans ask not what your country can do for you ask what you can do for your country.>
but was:
<and so my fellow Americans ask not what your country can do for you ask what you can do for your country>
```

An alternative would also be to update the expected output but it felt
better to keep the same expected output and disable flash-attention and
not just change the expected output to match the new behavior.
This commit is contained in:
Daniel Bevenius 2025-10-01 09:13:34 +02:00 committed by GitHub
parent 8c0855fd6b
commit 2a56869669
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
2 changed files with 5 additions and 2 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
bindings/java/src/main/java/io/github/ggerganov/whispercpp/params/WhisperContextParams.java
View File

@ -20,7 +20,7 @@ public class WhisperContextParams extends Structure {
/** Use GPU for inference (default = true) */
public CBool use_gpu;
/** Use flash attention (default = false) */
/** Use flash attention (default = true) */
public CBool flash_attn;
/** CUDA device to use (default = 0) */

5
bindings/java/src/test/java/io/github/ggerganov/whispercpp/WhisperCppTest.java
View File

@ -4,6 +4,7 @@ import static org.junit.jupiter.api.Assertions.*;
import io.github.ggerganov.whispercpp.bean.WhisperSegment;
import io.github.ggerganov.whispercpp.params.CBool;
import io.github.ggerganov.whispercpp.params.WhisperContextParams;
import io.github.ggerganov.whispercpp.params.WhisperFullParams;
import io.github.ggerganov.whispercpp.params.WhisperSamplingStrategy;
import org.junit.jupiter.api.BeforeAll;
@ -25,7 +26,9 @@ class WhisperCppTest {
//String modelName = "../../models/ggml-tiny.bin";
String modelName = "../../models/ggml-tiny.en.bin";
try {
whisper.initContext(modelName);
WhisperContextParams.ByValue contextParams = whisper.getContextDefaultParams();
contextParams.useFlashAttn(false); // Disable flash attention
whisper.initContext(modelName, contextParams);
//whisper.getFullDefaultParams(WhisperSamplingStrategy.WHISPER_SAMPLING_GREEDY);
//whisper.getJavaDefaultParams(WhisperSamplingStrategy.WHISPER_SAMPLING_BEAM_SEARCH);
modelInitialised = true;