Skip to content
Other

LPCombine-Qwen3.5-4B

Context combining for Qwen3.5

2 min readupdated

LPCombine-Qwen3.5-4B merges two independently captured contexts into a single resumable state. Built on the Qwen3.5-4B architecture by Simon Willani, the model uses a trained reader to propose tokens from the keys and values of the first full-attention layer and verifies candidates against the early cache. It is useful for developers working on efficient context management and caching in text-processing pipelines with the LatentPort library (a tool for working with model state caches).

Key features

  • Merges two independent contexts (left and right caches) into a single state
  • Supports BF16, NF4, and GGUF backends for cache operations
  • Uses a trained reader attached on top of a frozen base model
  • Passes equivalence tests against native token appending

What you can do

  • Call the combine() method to merge different contexts in Python using the LatentPort library
  • Load the model in NF4 or GGUF format to save VRAM when working with states
  • Run session queries to extract information from combined contexts

Example

import latentport as lp

with lp.load_model("latentport/LPCombine-Qwen3.5-4B", backend="nf4") as model:
    left = model.prefill("Nora belongs to team X.\n")
    right = model.prefill("Team X uses the passcode cedar.\n")
    merged = left.combine(right, model=model, method="cache")
    with model.session(merged) as session:
        print(session.ask("What is the passcode for Nora's team?").text)
Read moreDetails
Backends: BF16, NF4, and GGUF with token retention support
Training: FP32 reader trained on 18,432 synthetic tokens with a frozen base model
Limitations: Requires right-context replay and verification of recovered tokens before updating the session
License: Apache-2.0
Discuss Telegram 0 15

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/LPCombine-Qwen3.5-4B
    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/LPCombine-Qwen3.5-4B --local-dir ~/models/LPCombine-Qwen3.5-4B
    # The model is in ~/models. How to run it is in the model card (link above).
    Add more LoRAs to LPCombine-Qwen3.5-4B
  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. LPCombine-Qwen3.5-4B

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

latentport/LPCombine-Qwen3.5-4B
Author
latentport
Category
Other
Released
Posted

On NeyroLearn

Hugging Face

150

Ask AI about this model

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

Comments