tqcq/llama.cpp
0
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2025-07-23 19:25:51 +00:00
Code Issues Packages Projects Releases Wiki Activity

clip : cap max image size 1024 for qwen vl model (#13478)

Browse Source
This commit is contained in:
Xuan-Son Nguyen
2025-05-12 15:06:51 +02:00
committed by GitHub
parent 10d2af0eaa
commit de4c07f937
1 changed files with 8 additions and 4 deletions
Download Patch File Download Diff File Expand all files Collapse all files

12
tools/mtmd/clip.cpp
View File

@ -1909,16 +1909,20 @@ struct clip_model_loader {
} break;
case PROJECTOR_TYPE_QWEN2VL:
{
// max image size = sqrt(max_pixels)
// https://huggingface.co/Qwen/Qwen2-VL-7B-Instruct/blob/main/preprocessor_config.json
hparams.image_size = 3584;
// max image size = sqrt(max_pixels) = 3584
// ref: https://huggingface.co/Qwen/Qwen2-VL-7B-Instruct/blob/main/preprocessor_config.json
// however, the model use unreasonable memory past 1024 size, we force it to 1024 otherwise it's unusable
// ref: https://huggingface.co/Qwen/Qwen2-VL-2B-Instruct/discussions/10
hparams.image_size = 1024;
hparams.warmup_image_size = hparams.patch_size * 8;
} break;
case PROJECTOR_TYPE_QWEN25VL:
{
// max image size = sqrt(max_pixels)
// https://huggingface.co/Qwen/Qwen2.5-VL-7B-Instruct/blob/main/preprocessor_config.json
hparams.image_size = 3584;
// however, the model use unreasonable memory past 1024 size, we force it to 1024 otherwise it's unusable
// ref: https://huggingface.co/Qwen/Qwen2-VL-2B-Instruct/discussions/10
hparams.image_size = 1024;
hparams.warmup_image_size = hparams.patch_size * 8;
get_u32(KEY_WIN_ATTN_PATTERN, hparams.n_wa_pattern);
} break;