Aditya02/IndicF5

1## Backbones quick introduction2 3 4### unett.py5- flat unet transformer6- structure same as in e2-tts & voicebox paper except using rotary pos emb7- update: allow possible abs pos emb & convnextv2 blocks for embedded text before concat8 9### dit.py10- adaln-zero dit11- embedded timestep as condition12- concatted noised_input + masked_cond + embedded_text, linear proj in13- possible abs pos emb & convnextv2 blocks for embedded text before concat14- possible long skip connection (first layer to last layer)15 16### mmdit.py17- sd3 structure18- timestep as condition19- left stream: text embedded and applied a abs pos emb20- right stream: masked_cond & noised_input concatted and with same conv pos emb as unett21