Skip to content
Other

ResNet-18 Tensor-Train

model compression for dysplasia grading

2 min readupdated

This model is a compressed version of PyTorch-based ResNet-18, featuring 3x3 convolutional kernels decomposed using the Tensor-Train format. It is trained to grade the severity of oral epithelial dysplasia on the OralEpitheliumDB dataset. The main feature of this method is that reducing the parameter count does not decrease the computational volume (GFLOPs): the kernels are reassembled on each forward pass, which cuts memory requirements while increasing latency. The model is intended for research and educational tasks in medical image analysis.

Key features

  • Significant parameter reduction (down to 4.60 M compared to 11.18 M for the baseline).
  • Preservation of computational complexity (GFLOPs remained at 3.63).
  • Use of the Tensor-Train method with four cores per single convolutional kernel.
  • Efficiency comparison with alternatives such as the Tucker-2 model.

What you can do

  • Apply the model for tissue histological image analysis and classification in research projects.
  • Conduct experiments to evaluate latency and resource consumption on CPU and GPU.
  • Compare the performance quality of Tensor-Train with the baseline ResNet-18 architecture and the Tucker-2 variant.

Example

# pip install torch torchvision safetensors huggingface_hub
from huggingface_hub import hf_hub_download
import importlib.util, sys
from PIL import Image

path = hf_hub_download("LuisPaim/oed-resnet18-tensortrain", "oed_models.py")
spec = importlib.util.spec_from_file_location("oed_models", path)
oed = importlib.util.module_from_spec(spec); sys.modules["oed_models"] = oed; spec.loader.exec_module(oed)

model = oed.from_pretrained("LuisPaim/oed-resnet18-tensortrain")
print(oed.predict(model, Image.open("roi.tif")))
Read moreDetails
Parameters: 4.60 M (59% reduction compared to baseline)
GFLOPs: 3.63 (unchanged compared to baseline)
Metrics (on a test set of 92 images): AUC 0.9787, F1 0.8896, QWK 0.9195, Accuracy 0.8913
Hardware and latency: GPU latency 7.01 ms (RTX 3050), CPU latency 52.2 ms (Intel Xeon)
Limitations: for research and education only, not a medical device
Discuss Telegram 0

Will it run?

Runs in F32

Fits without quantisation.

  • F32weights 0.0 GB · ready files0.5 GB
  • BF16weights 0.0 GB0.5 GB
  • FP8 / INT8recommendedweights 0.0 GB0.5 GB
  • 4-bitweights 0.0 GB0.5 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 · F32 — from RTX 2060

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.

    # LuisPaim/oed-resnet18-tensortrain
    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 LuisPaim/oed-resnet18-tensortrain --local-dir ~/models/oed-resnet18-tensortrain
    # The model is in ~/models. How to run it is in the model card (link above).
  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

New posts about versions of oed-resnet18 (quants, finetunes, LoRAs) will show up under the bell.

Fine-tunes 2All on Hugging Face

About the model

LuisPaim/oed-resnet18-tensortrain
Author
LuisPaim
Category
Other
Released
Posted

On NeyroLearn

Hugging Face

00

Ask AI about this model

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

Comments