Skip to content
Other

LPCompact-Qwen3.5-9B

attention cache compactor for long contexts

1 min readupdated

This compact 393,473-parameter selector compresses the attention cache of the base Qwen/Qwen3.5-9B model. By scoring position importance from attention values across specific layers, it retains critical keys and values alongside recurrence and convolution states. This lets you handle long contexts far more efficiently without significant accuracy loss—ideal for developers and researchers working with large text volumes and extended sessions.

Key features

  • Cache compression: reduces the memory footprint needed to store long-context history.
  • Flexible policy: the default 8x policy keeps the first and last 512 positions, with a minimum floor of 1,024 positions.
  • Tunable compression: set the ratio parameter to save even more memory.
  • Broad compatibility: supports BF16, NF4, and compatible GGUF weights.

What you can do

  • Integrate into pipelines: use the latentport library for prefill and generation sessions with a reduced cache.
  • Process large documents: run search tasks across code repositories and structured data over long histories.

Example

import latentport as lp

with lp.load_model("Qwen/Qwen3.5-9B", backend="bf16") as model:
    state = model.prefill(history)
    child = state.compact()
    answer = model.session(child).ask("Continue the task.")
Read moreDetails
Requires LatentPort 0.4.0 or newer.
Compression starts at 65,536 physical positions by default.
Training included a warmup phase on 16 histories with 256 AdamW updates, followed by main training on 32 histories.
Discuss Telegram 0 35

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.

    # latentport/LPCompact-Qwen3.5-9B
    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 latentport/LPCompact-Qwen3.5-9B --local-dir ~/models/LPCompact-Qwen3.5-9B
    # The model is in ~/models. How to run it is in the model card (link above).
    Add more LoRAs to LPCompact-Qwen3.5-9B
  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. Qwen3.5-9B-Basefamily root
  2. Qwen3.5-9B
  3. LPCompact-Qwen3.5-9B

New posts about versions of Qwen3.5-9B-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 6All on Hugging Face
Merges 1All on Hugging Face

About the model

latentport/LPCompact-Qwen3.5-9B
Author
latentport
Category
Other
Released
Posted

On NeyroLearn

Hugging Face

350

Ask AI about this model

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

Comments

Computer vision

VisAlloc-9B

VisAlloc-9B