Skip to content
Other

HEGM-VK-LSVD-K12

short video watch time prediction

2 min read

A pretrained checkpoint of the HEGM (Hierarchical Exponential-Gaussian Mixtures) architecture for short video recommender systems. The model predicts the watch time distribution of a clip: the exponential component handles quick swipes, while a mixture of 12 Gaussian (normal) distributions handles engaged viewing, combining them via a trainable skip gate. The checkpoint was trained on the up0.01_ir0.01 subset of the VK-LSVD dataset and will be useful for researchers and developers of recommendation algorithms.

Key features

  • Separates quick skips and engaged views using a trainable skip gate and 12 Gaussian components.
  • Includes a ready-to-use PyTorch Lightning checkpoint (last.ckpt) and the original experiment config config.yaml.
  • The series also includes variants with different numbers of components: K=3, K=6, and K=9.
  • Custom PyTorch architecture: does not load directly via standard transformers.AutoModel.from_pretrained.

What you can do

  • Evaluate watch time prediction quality on the VK-LSVD dataset test split and reproduce paper results.
  • Compare prediction accuracy across different mixture component counts (matching K12 against K3, K6, and K9 versions).
  • Use as a base model or baseline when developing ranking and recommendations for short video feeds.

Example

git clone [https://github.com/rw404/HEGM.git](https://github.com/rw404/HEGM.git)
cd HEGM
pip install -r requirements.txt
python scripts/download_artifacts.py --splits
python evaluate.py --config configs/vklsvd_k12.yaml --checkpoint checkpoints/vklsvd_k12/last.ckpt
Read moreDetails
Metrics on the VK-LSVD test set: MSE — 479.6837 sec², MAE — 15.7679 sec, XAUC — 0.6202.
Running evaluation requires processed data splits (including meta.json) from the official repository.
Dataset paths in config.yaml are specified relative to the GitHub project root.
Discuss Telegram 0 16

Will it run?

Runs in BF16

Fits without quantisation.

  • BF16weights 0.0 GB · ready files0.5 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 · BF16 — 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.

    # glvsks/HEGM-VK-LSVD-K12
    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 glvsks/HEGM-VK-LSVD-K12 --local-dir ~/models/HEGM-VK-LSVD-K12
    # The model is in ~/models. How to run it is in the model card (link above).
    Add more LoRAs to HEGM-VK-LSVD-K12
  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 HEGM-VK-LSVD-K12 (quants, finetunes, LoRAs) will show up under the bell.

No known relatives of this model yet. All derivatives on Hugging Face

About the model

glvsks/HEGM-VK-LSVD-K12
Author
glvsks
Category
Other
Released
Posted

On NeyroLearn

Hugging Face

160

Ask AI about this model

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

Comments

Other

K-ALUAN

K-ALUAN