Hannes Fassold comments

Results 7 comments of


                                            Hannes Fassold

[Feature request] AWQ (activation-aware weight quantization) 4-bit quantization

No, paper doesn't mention retraining, as far as I could understand it Also in the code I cannot find any training loop. Main function for generating quantized model seems to...

[Feature request] AWQ (activation-aware weight quantization) 4-bit quantization

@WesCook Thanks, SpQR looks also interesting. Although AWQ seems to be the 'easier' format (to understand and implement). Just from a first look at both papers.

Quantization support needed for Ovis2.5 9B

use the bitsandbytes 4-byte quantization instead

Openmmlab website certificate expired

See https://github.com/open-mmlab/mmcv/issues/3302#issuecomment-3214056533

[Bug] SSL certificate expired for https://download.openmmlab.com

Thank you @gboeer !!!

Inference without Flash Attention is Needed!

You can build flashattn packe from source

Will Ovis2.5 be officially supported by Transformers, and can it be fine-tuned like Qwen2.5-VL?

Ovis2.5 runs fine with recent Transformer versions It should NOT be included (like Ovis-2) in the 'officiall' transformers package (see https://github.com/AIDC-AI/Ovis/issues/124 )