Skip to content
Computer vision

TR-OCR-Markdown v0.4

LoRA for DeepSeek-OCR-2

2 min readupdated

Built on unsloth/DeepSeek-OCR-2 (3.5B parameters), this updated adapter (v0.4) turns images of Turkish documents—forms, contracts, laws, receipts, and tables—into structured Markdown while preserving headings, lists, tables, and LaTeX formulas. Trained on a mix of synthetic and real-world data, it substantially outperforms earlier versions on both real scans and digital PDFs, driving down the character error rate (CER).

Key features

  • Built for Turkish: Accurately extracts text from complex documents and printed materials.
  • Markdown formatting: Automatically detects and formats headings, lists, tables, and LaTeX formulas.
  • Lightweight footprint: An 86M-parameter LoRA adapter (346 MB) that leaves the base vision encoders untouched.
  • Lower error rates: Major CER improvements over v0.3 across real-world test sets.

What you can do

  • Plug it into document-processing pipelines to convert scanned laws and reports into clean Markdown.
  • Parse receipts, waybills, and invoices using a dedicated plain-text or Markdown prompt.
  • Digitize old books, academic papers, and official gazettes.
  • Run it in local inference scripts on GPUs with bf16 support.

Example

import torch
from transformers import AutoModel, AutoTokenizer
from peft import PeftModel

BASE, REPO = "unsloth/DeepSeek-OCR-2", "msbayindir/tr-ocr-markdown-deepseek-ocr2-lora-v0.4"
tok = AutoTokenizer.from_pretrained(BASE, trust_remote_code=True)
base = AutoModel.from_pretrained(BASE, trust_remote_code=True, use_safetensors=True, _attn_implementation="eager",
                                 torch_dtype=torch.bfloat16).eval().cuda()
model = PeftModel.from_pretrained(base, REPO).eval()
model = model.merge_and_unload()

md = model.infer(tok, prompt="\nFree. OCR. ", image_file="sayfa.png", output_path="/tmp/out",
                 base_size=1024, image_size=768, crop_mode=True)
print(md)
Read moreDetails
• Base model: unsloth/DeepSeek-OCR-2 (3.5B parameters, Apache-2.0).
• Training: 1 epoch, LoRA r=16, 20.5k examples.
• Metrics (CER): Dropped to 0.4% on born-digital laws and 25.4% on scanned Resmî Gazete issues.
• Limitations: Human review is mandatory for legal and official use cases.
Discuss Telegram 0 33

Will it run?

Runs in BF16

Fits without quantisation.

  • BF16weights 6.6 GB · ready files7.8 GB
  • FP8 / INT8recommendedweights 3.5 GB4.3 GB
  • 4-bitweights 2.1 GB2.8 GB

Tip: go 8-bit.FP8 / INT8 is within about a percent of full precision in quality, takes half the memory and runs faster on RTX 40/50 and H100. Full precision only pays off for fine-tuning.

What to get for this model

Massed ComputeOpen Massed Compute

Your own GPU: FP8 / INT8 — from RTX 2060 · BF16 — from RTX 5060 Ti 8GB

Massed Compute hourly prices, November 2026. The discounted price applies with the coupon at checkout.

This is a LoRA: counted together with its base model DeepSeek-OCR-2.

Estimate: weights plus working memory. Long videos, high resolutions and long contexts need more.

Run it on Massed Compute

The script downloads the model to the rented machine. How to run it is in the model card.

  1. 1. Rent a machine with a fitting GPU

    Open Massed Compute
  2. 2. Paste the script into the machine's terminal

    Connect over SSH (the Massed Compute panel shows the command) or open a terminal on the machine's desktop.

    HF_TOKEN / CIVITAI_TOKENThe script starts with token lines: paste yours between the quotes. The Hugging Face token is for gated models, the Civitai one for Civitai models.

    # msbayindir/tr-ocr-markdown-deepseek-ocr2-lora-v0.4
    # ===== Tokens: paste yours between the quotes =====
    export HF_TOKEN=""        # huggingface.co/settings/tokens, for gated models
    # ==========================================
    mkdir -p ~/models
    curl -LsSf https://astral.sh/uv/install.sh | sh
    source "$HOME/.local/bin/env"
    uv tool install "huggingface_hub[cli]"
    hf download unsloth/DeepSeek-OCR-2 --local-dir ~/models/DeepSeek-OCR-2
    hf download msbayindir/tr-ocr-markdown-deepseek-ocr2-lora-v0.4 --local-dir ~/models/tr-ocr-markdown-deepseek-ocr2-lora-v0.4
    # The model is in ~/models. How to run it is in the model card (link above).
    Add more LoRAs to DeepSeek-OCR-2
  3. 3. Use it

    The model is in ~/models. How to run it is in the model card (link above).

Stop the machine in the panel when you are done: it bills by the minute while it runs.

How to speed it up

Picked for this model type. Choose your hardware above to see what fits you exactly.

  • FlashAttention 44.0.0b40from November 18, 2026new release

    fastest exact attention on B200

    Kernels for Blackwell (beta). Exact attention without quantisation — fine for everything, LLMs included.

    How to enable: pip install flash-attn-4; vLLM and SGLang pick it up themselves.

    Earlier generations (2)
    • FlashAttention 3 — The Hopper (H100/H200) version with FP8.
    • FlashAttention 2 — The standard for RTX 30/40/50 and A100: exact attention without the memory cost of long contexts.
  • torch.compile2.15.0from October 28, 2026new release

    10–30% for free

    Compiles the model for your GPU. Slower first run, faster afterwards.

    How to enable: model = torch.compile(model); in ComfyUI the TorchCompileModel node.

Model family

Requires the base model

unsloth/DeepSeek-OCR-2

This model is a LoRA / adapter. Download the base model first, then load this one on top.

  1. DeepSeek-OCR-2family root
  2. DeepSeek-OCR-2
  3. tr-ocr-markdown-deepseek-ocr2-lora-v0.4

New posts about versions of DeepSeek-OCR-2 (quants, finetunes, LoRAs) will show up under the bell.

Quantised versions 12All on Hugging Face
Fine-tunes 12All on Hugging Face
LoRAs and adapters 9All on Hugging Face
Merges 1All on Hugging Face

About the model

msbayindir/tr-ocr-markdown-deepseek-ocr2-lora-v0.4
Author
msbayindir
Category
Computer vision
Released
Posted

On NeyroLearn

Hugging Face

330

Ask AI about this model

Answered by Gemini 4 · the chat is kept only in this tab

Comments