Aller au contenu
ModelPrésentéExterne
Télécharger

Nyansapo Ocr Trocr LoRA v1 0 2

maxtargetlen reduced from 128 → 4 (root fix for hallucination) norepeatngramsize set to 0 (was hurting short outputs) lengthpenalty set to 1.0 (was 2.0 — was pushing model to generate more tokens)

TypeLoRAModèle de basemicrosoft/trocr-base-handwrittenVersionadapter_modelCClasse

Source du modèle

Type
LoRA
Taille du fichier
19,2 MB
Modèle de base
microsoft/trocr-base-handwritten
Version
vadapter_model
Mis à jour
18 sept. 2026
Téléchargements
2
Source
Hugging Face

Description de la source

Nyansapo OCR — TrOCR LoRA v1.0.1

What changed from v1.0.0

  • max_target_len reduced from 128 → 4 (root fix for hallucination)
  • no_repeat_ngram_size set to 0 (was hurting short outputs)
  • length_penalty set to 1.0 (was 2.0 — was pushing model to generate more tokens)

Training details

  • Base model: microsoft/trocr-base-handwritten
  • PEFT: LoRA (r=16, alpha=32)
  • Dataset: Nzyoka19/ocr_images

Sources

1 source
Hugging FacePrincipalVersion e8c954972d9e45a7edd73306eabf34001c459010
huggingface.co/Nzyoka19/nyansapo-ocr-trocr-lora_v1-0-2/tree/e8c954972d9e45a7edd73306eabf34001c459010

Vérifié 3 sept.

Artefacts du modèle

1 artefact

adapter_model.safetensors

safetensors · 19,2 MB · SHA-256 129429aaadc8…87f6 · Hugging Face

Download file

Extraits de sources

2 extraits

--- base_model: microsoft/trocr-base-handwritten tags: [ocr, lora, peft, trocr, image-to-text, nyansapo] license: mit --- # Nyansapo OCR — TrOCR LoRA v1.0.1 ## What changed from v1.0.0 - `max_target_len` reduced from 128 → 4 (root fix for hallucination) - `no_repeat_ngram_size` set to 0 (was hurting short outputs) - `length_penalty` set to 1.0 (was 2.0 — was pushing model to generate more tokens) ## Training details - **Base model:** microsoft/trocr-base-handwritten - **PEFT:** LoRA (r=16, alpha=32) - **Dataset:** Nzyoka19/ocr_images

Source context: 2 downloads · 0 likes · Pipeline image-to-text · Library peft · Repo Nzyoka19/nyansapo-ocr-trocr-lora_v1-0-2