This is the <a target="_blank" rel="ugc" href="https://civarchive.com/articles/36274/converting-a-2b-anima-model-to-a-29b-anima-model">converted to ANIMA 2.9B </a>version of my <a target="_blank" rel="ugc" href="https://civarchive.com/models/2632277/anima-v1-turbo-aio">ANIMA-V1-Turbo-AIO</a>.For now, only V5, if a previous version is requested, i'll see what i can do ;-)Now also available as INT8 ConvRot. Conversion done this way:<pre><code>convert-to-quant -i ANIMA-29B-Turbo-AIO_V5.safetensors -o ANIMA-29B-Turbo-AIO_V5-INT8.safetensors --int8 --convrot --convrot-group-size 256 --scaling_mode row --comfy_quant --save-quant-metadata --simple --anima --exclude-layers '(first_stage_model|cond_stage_model)'</code></pre>A quick generation and comparaison between two pictures generated using BF16 and INT8 convrot yielded only very small differences.<img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/0febc5b7-cbf6-43a8-ac8a-756ab04641a7/original=true/0febc5b7-cbf6-43a8-ac8a-756ab04641a7.jpeg" />And it is actually faster to generate using INT8_convrot!!! (2.3s/it vs 1.4s/it on my RTX 3060).

ANIMA-2.9B-Turbo-AIO

Posted on Civitai, upload found elsewhere are FORBIDDEN and done without the author permission.

anima29BTurboAIO_v5_bf16.safetensors

anima29BTurboAIO_v5_int8.safetensors

This is the <a target="_blank" rel="ugc" href="https://civitai.com/articles/36274/converting-a-2b-anima-model-to-a-29b-anima-model">converted to ANIMA 2.9B </a>version of my <a target="_blank" rel="ugc" href="https://civitai.com/models/2632277/anima-v1-turbo-aio">ANIMA-V1-Turbo-AIO</a>.For now, only V5, if a previous version is requested, i'll see what i can do ;-)<s>I'll provide an INT8_convrot version asap too.</s> Upload is on-going. Conversion done this way:convert-to-quant -i ANIMA-29B-Turbo-AIO_V5.safetensors -o ANIMA-29B-Turbo-AIO_V5-INT8.safetensors --int8 --convrot --convrot-group-size 256 --scaling_mode row --comfy_quant --save-quant-metadata --simple --anima --exclude-layers '(first_stage_model|cond_stage_model)'A quick generation and comparaison between two pictures generated using BF16 and INT8 convrot yielded only very small differences.And it is actually faster to generate using INT8_convrot!!! (2.3s/it vs 1.4s/it on my RTX 3060).

Files with SHA256: d0c8550d7dc49952aa8685275947e712a54ca682e81b03c8da0540dd33cdd216