Skip to content
Computer vision

peshtur

A fast visual choice model with integrated OCR

2 min readupdated

The peshtur model is built on Qwen3.5-4B (a base model with 4.54 billion parameters) and handles multiple-choice visual question-answering tasks. Instead of generating text token by token, the model outputs a calibrated probability distribution across all options (from A to Z) in a single forward pass. The system features a built-in lightweight OCR module (PP-OCRv5) to recognize text in images right during the request. This development will be useful for developers and researchers who need high processing speed for visual data along with accuracy in markup tasks and the analysis of interfaces or documents.

Key features

  • Direct output: Evaluates probabilities of answer options in a single forward pass without token generation.
  • Integrated OCR: Automatically recognizes text in images using PP-OCRv5 on every request.
  • Mixed architecture: Combines full attention layers (SDPA) with Gated DeltaNet recurrent layers.
  • Efficiency: Fine-tuned using LoRA across all linear matrices of the language component.

What you can do

  • Document and receipt analysis: Apply it to extract key information from forms, receipts, and infographics.
  • Content moderation: Use the model to classify memes and check images for policy violations.
  • UI tasks: Process screenshots of web interfaces and answer questions about element positioning.

Example

from peshtur import Engine

engine = Engine()
response, _ = engine(
    {},
    {"answer": {"type": "choice", "instructions": "Which object is closer to the camera?",
                "criteria": {"A": "the bicycle", "B": "the bench"}}},
    images=["street.jpg"],
)
print(response["answers"]["answer"])
Read moreDetails
Benchmarks: 75.44 on the full public Decision Index Vision 0.3 dataset (12,210 rows)
Hardware requirements: GPU with 16 GB VRAM or more (peak consumption around 12.53 GB)
Limitations: R-Bench-M (33.38) and MMMU-Pro vision (41.21) are constrained by the 4B base capabilities
Training details: 1 epoch on 323,264 examples with distillation from three 27B teacher models
Discuss Telegram 0 14

Will it run?

Runs in BF16

Fits without quantisation.

  • BF16weights 8.5 GB · ready files9.8 GB
  • FP8 / INT8recommendedweights 4.2 GB5.2 GB
  • 4-bitweights 2.4 GB3.1 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 3080

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

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.

    # plantwaterer/peshtur
    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 plantwaterer/peshtur --local-dir ~/models/peshtur
    # The model is in ~/models. How to run it is in the model card (link above).
    Add more LoRAs to peshtur
  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

  1. Qwen3.5-4B-Basefamily root
  2. Qwen3.5-4B
  3. peshtur

New posts about versions of Qwen3.5-4B-Base (quants, finetunes, LoRAs) will show up under the bell.

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

About the model

plantwaterer/peshtur
Author
plantwaterer
Category
Computer vision
Released
Posted

On NeyroLearn

Hugging Face

140

Ask AI about this model

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

Comments