Andrii Staikov

icon indicating an email [email protected]

icon company OpenVINO engineer at Intel icon location Bottom of the Pacific Ocean icon location Pretending to be a software developer

Results 12 issues of Andrii Staikov

Move SDPAToPA-related functionality to runtime

4 comment

## Description Move the SDPAToPA-related functionality to the transformation to avoid decoupling the code that needs to be a part of a single component. This would help to make the...

category: continuous batching
category: speculative decoding
category: GHA

[TRANSFORMATIONS] RoPE support different dimensions options for flux.1-schnell

### Details: The flux.1-schnell model has changes after updating diffusers from 0.33.1 to 0.35.2 which caused another dimensions arrangement for RoPE in the model: [batch, num_heads, -1, head_size] -> [batch,...

category: transformations