tqcq/llama.cpp
0
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2025-07-14 22:58:10 +00:00
Code Issues Packages Projects Releases Wiki Activity

Merge branch 'master' into gg/llama-kv-cache

Browse Source
This commit is contained in:
Georgi Gerganov
2025-02-21 19:14:07 +02:00
parent 3753b30d65 51f311e057
commit c4c0a4d13c
23 changed files with 810 additions and 29 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
examples/llava/clip.cpp
View File

@ -2712,9 +2712,13 @@ bool clip_image_batch_encode(clip_ctx * ctx, const int n_threads, const clip_ima
if (!ctx->has_glm_projector) {
struct ggml_tensor * patches = ggml_graph_get_tensor(gf, "patches");
// The patches vector is used to get rows to index into the embeds with;
// we should skip dim 0 only if we have CLS to avoid going out of bounds
// when retrieving the rows.
int patch_offset = ctx->has_class_embedding ? 1 : 0;
int* patches_data = (int*)malloc(ggml_nbytes(patches));
for (int i = 0; i < num_patches; i++) {
patches_data[i] = i + 1;
patches_data[i] = i + patch_offset;
}
ggml_backend_tensor_set(patches, patches_data, 0, ggml_nbytes(patches));
free(patches_data);

BIN
examples/server/public/index.html.gz
View File

Binary file not shown.

1
examples/server/webui/src/components/ChatScreen.tsx
View File

@ -228,6 +228,7 @@ export default function ChatScreen() {
value={inputMsg}
onChange={(e) => setInputMsg(e.target.value)}
onKeyDown={(e) => {
if (e.nativeEvent.isComposing || e.keyCode === 229) return;
if (e.key === 'Enter' && e.shiftKey) return;
if (e.key === 'Enter' && !e.shiftKey) {
e.preventDefault();

2
examples/server/webui/src/utils/llama-vscode.ts
View File

@ -40,7 +40,7 @@ export const useVSCodeContext = (
window.addEventListener('message', handleMessage);
return () => window.removeEventListener('message', handleMessage);
}, []);
}, [inputRef, setInputMsg]);
// Add a keydown listener that sends the "escapePressed" message to the parent window
useEffect(() => {