MagicSource

Hi, wanna feedback some visual issues on different languals. I hope your guys take a look on hopefully resolve it. Here is a render result: As you can see, there...

I dnk why this isn't a default option, when long press blank area at editor inside, no paste menu show up. When selected some words, it shows. Not sure how...

[BUG] - Sometimes crash on macOS

2

When inputting, sometimes will crash (perhaps with probability of 40%). input with Chinese specifically. Any chances to have a test? ``` Thread 0 Crashed:: (input method 18934 com.apple.inputmethod.SCIM) Dispatch queue:...

would consider add macos support? (or maybe CPU)

Qwen2.5 VL dataset sampler based on modality length

2

Currently since Qwen2.5 VL with various image sizes, when training on relatively large image size, the image tokens are various for each batch, the default resampling logic would cause OOM...

Any downstream image matching applications?

1

thanks for this wonderful project, the image matching looks more and more faster and accurate, any other down stream iamge matching project could be attached? for example: - localization; -...

I found why using ratio of 0.7, the duration didn't changed while speech got faster. After look carefully, the saved wav have about 1s silence, why ?

Siglip2 model support request

2

Would consider support Siglip2 model as well?

Hi, will exl2 quantization support

Hi, i had trained a llava like MLLM but with a customized vision encoder (siglip2) Am wondering if there any steps to follow to create a gguf / uqff format...