upshift/infomaniak-translategemma-27b-it-W4A16
Image-Text-to-Text • 27B • Updated • 117
TranslateGemma
4-bit (W4A16) builds of Infomaniak's vLLM TranslateGemma 4B/12B/27B, GPTQ and AWQ, with chrF++ on FLORES+ and Tatoeba.
Note W4A16 GPTQ · 16.9 GB · chrF++ FLORES+ 62.4 / Tatoeba 70.0 (bf16 62.5 / 70.3)
upshift/infomaniak-translategemma-12b-it-W4A16
Image-Text-to-Text • 12B • Updated • 106Note W4A16 GPTQ · 8.4 GB · chrF++ FLORES+ 60.9 / Tatoeba 70.3 (bf16 61.5 / 70.8)
upshift/infomaniak-translategemma-4b-it-W4A16
Image-Text-to-Text • 4B • Updated • 18Note W4A16 GPTQ · 3.8 GB · chrF++ FLORES+ 58.8 / Tatoeba 65.2 (bf16 59.3 / 65.8)
upshift/infomaniak-translategemma-27b-it-W4A16-awq
Image-Text-to-Text • 27B • Updated • 83Note W4A16 AWQ · 16.9 GB · chrF++ FLORES+ 60.5 / Tatoeba 68.2 (bf16 62.5 / 70.3)
upshift/infomaniak-translategemma-12b-it-W4A16-awq
Image-Text-to-Text • 12B • Updated • 45Note W4A16 AWQ · 8.4 GB · chrF++ FLORES+ 60.2 / Tatoeba 69.4 (bf16 61.5 / 70.8)
upshift/infomaniak-translategemma-4b-it-W4A16-awq
Image-Text-to-Text • 4B • Updated • 27Note W4A16 AWQ · 3.8 GB · chrF++ FLORES+ 57.4 / Tatoeba 65.3 (bf16 59.3 / 65.8)