Skip to content
Other

HGRN-1.3B-semi-structured-EHWS-80pct-seed2

Sparse Version of a Language Model

1 min readupdated

This model is an experimental version of the base fla-hub/hgrn-1.3B-100B, optimized using a two-phase ADMM algorithm to achieve 80% semi-structured sparsity. The model's linear layers retain an equal number of weights in each row, while the embeddings and final projection remain dense. The model is intended for researchers and developers exploring efficient architectures with linear attention and sparse weights. Running it requires the flash-linear-attention library.

Key features

  • Semi-structured sparsity: 80% of the weights in linear layers are removed while maintaining a uniform structure across rows.
  • ADMM optimization: Two-phase training with C4 data calibration to minimize accuracy loss.
  • Efficiency: Training took just 3.71 hours on a single A100 40GB accelerator.
  • Compatibility: Full integration with the standard AutoModelForCausalLM interface via the fla library.

What you can do

  • Compression research: Test the behavior of sparse linear models in text generation tasks.
  • Benchmarking: Compare performance and perplexity metrics against the base dense model on datasets like WikiText-2.
  • Hardware experiments: Evaluate the potential for compute acceleration through uniform layer sparsity.

Example

from transformers import AutoModelForCausalLM
import fla

model = AutoModelForCausalLM.from_pretrained(
    "ikimyaii/HGRN-1.3B-semi-structured-EHWS-80pct-seed2",
    device_map="auto"
)
print(model)
Read moreDetails
Perplexity: WikiText-2 (42.49), C4 (29.72)
Average zero-shot (lm-evaluation-harness): 0.3821
Base model: fla-hub/hgrn-1.3B-100B
Parameters: 1.36 billion
Discuss Telegram 0 18

Will it run?

Runs in F32

Fits without quantisation.

  • F32weights 5.1 GB · ready files6.1 GB
  • BF16weights 2.5 GB3.3 GB
  • FP8 / INT8recommendedweights 1.3 GB1.9 GB
  • 4-bitweights 0.7 GB1.3 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 5060 Ti 8GB

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.

    # ikimyaii/HGRN-1.3B-semi-structured-EHWS-80pct-seed2
    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 ikimyaii/HGRN-1.3B-semi-structured-EHWS-80pct-seed2 --local-dir ~/models/HGRN-1.3B-semi-structured-EHWS-80pct-seed2
    # 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 hgrn-1.3B-100B (quants, finetunes, LoRAs) will show up under the bell.

Fine-tunes 12All on Hugging Face

About the model

ikimyaii/HGRN-1.3B-semi-structured-EHWS-80pct-seed2
Author
ikimyaii
Category
Other
Released
Posted

On NeyroLearn

Hugging Face

180

Ask AI about this model

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

Comments