Florence-2 Large
Free.ai (self-hosted)
·
ocr
·
~300 Ii-token page
Florence-2 Large yi an Imodeli ye-OCR eyenziwe ngu Microsoft. Enamandla kakhulu kwi Complex document understanding, region-grounded OCR.. I-self-hosted kwi-Free.ai GPUs — iqhuba ngokukhululekileyo kwi-token pool yakho yosuku (300 tokens Iphepha ngalinye). Ikhutshwe phantsi kwe MIT — ukusetyenziswa korhwebo kuvunyelwe kwi Free.ai.
Sebenzisa nge-API
I-REST API ehambelana ne-OpenAI. Yenza iqhosha kwaye unxulumane nale modeli kwimizuzu.
curl -X POST https://api.free.ai/v1/ocr/ \
-H "Authorization: Bearer sk-free-..." \
-H "Content-Type: application/json" \
-d '{"model":"florence-2","prompt":"your prompt here"}'Imibuzo ebuzwa rhoqo
Florence-2 Large ikhupha umbhalo ukusuka kumfanekiso kunye ne-PDF. Layisha phezulu iphepha elishicilelwe, imifanekiso yekhusi, okanye umfanekiso woxwebhu kwaye Florence-2 Large ibuyisela umbhalo — igcina ukuphela komgca kunye (kwi-engine ezikwaziyo) itephu, ifomula, kunye nokwakhiwa.
Florence-2 Large iphatha amaqela amaninzi e-scripts - Latin, Cyrillic, CJK, Arabic, Devanagari, kunye nezinye. I / ocr / i landing ine uluhlu olupheleleyo lwesiNgesi ngenjini.
Ewe, ukubhala ngesandla okushicilelweyo okucocekileyo; i-cursive ne-shorthand zinzima. TrOCR kunye ne-transformer-based engines yexesha elizayo zingcono kakhulu kune-Tesseract ekhoyo ngoku kwi-writing.
Iinjini eziziqondayo i-Layout (GOT-OCR, Marker, Florence-2) zibuyisela iMarkdown/iiteyibhile ze-HTML ne-LaTeX zefomula. Iinjini zombhalo ocacileyo zibuyisela umbhalo ongenasakhiwo — khangela indlela yemveliso yemodeli.
Florence-2 Large iqhuba kwiGPUs zethu ukusuka kwi-pool yakho yasimahla yosuku; $5 → 200,000 iitokeni ezihlawulwe emva. ~300 iitokeni ngalinye.
PNG, JPG, WebP, HEIC, BMP, kunye neePDF ezine- kunye nee-multi-page ukuya kwi-50 MB. Amaphepha ajikelezayo okanye ajikelezayo alungiswe ngokuzenzekelayo phambi kokukhupha.
Umgangatho wophawu uhlala>98% kwiphepha elishicilelwe kakuhle, liwela kwimifanekiso ephantsi-res, i-scans ethe tyaba kakhulu, okanye ifonti ezingaqhelekanga. Thelekisa iinjini kwi /ocr/compare/ xa umgangatho ubalulekile.
Ewe — /batch/ ivuma isiqulathi seefayili semifanekiso/PDFs. Ukutywina ngalunye kuwela kwi /account/?tab=history. Ukugcina umthi wesiqulathi seefayili sebenzisa i-API.
Ewe — UTHENGA ifayili yakho ku /v1/ocr/ ngemodeli "Florence-2 Large". Ibuyisela i-JSON ngombhalo + (xa ixhaswe) ubeko + ibhokisi yomimandla yegama-level. /api/ inesiqendu esipheleleyo.
Iimodeli ezimkelweyo zigcina ifayili yakho kwi-GPU yethu; i-premium idlula nge-DPA. Amaxwebhu asuswa emva kwefestile yokwabelana. Asiqeqeshi kwingxelo yakho.
Ewe - Free.ai inikezela ngenkonzo yorhwebo lombhalo okhutshwe. Ufuna iimvume zomthombo woxwebhu.
1-5 imizuzwana ngaphaya kwephepha ngalinye. Iinjini eziziqondayo i-layout (iiteyibhile, i-mathematics) zithatha ixesha elide. I-PDFs eneephepha ezininzi ziyi-scale ngokuthe ngqo.