Skip to content
Other

Rambley-V3-50M

an ultra-compact experimental language model

1 min readupdated

Created by glossia, Rambley-V3-50M is a 50-million-parameter language model (LLM) trained from scratch on roughly 36B tokens. The architecture packs in a range of cutting-edge techniques, including Virtual Head Attention, Sliding Window Attention (SWA), HoPE positional embeddings, and the Muon optimizer. Designed for research and experimentation with tiny model architectures, it brings together modern industry innovations at a miniature scale.

Key features

  • Trained natively in BF16 for maximum stability.
  • Uses modern Muon and AdamW optimizers paired with an AIR LR scheduler.
  • Leverages Virtual Head Attention and headwise gated attention to keep parameter counts low.
  • Incorporates Identity Hyperconnections and learned attention sinks.

What you can do

  • Run the pre-training codebase to study custom architectural pipelines.
  • Explore how ultra-small LLMs behave when equipped with advanced attention and optimization mechanisms.

Example

The model card does not include a ready-to-run inference script, though you can run the model with standard Hugging Face libraries alongside the author's pre-training repository.

Read moreDetails
Parameters: ~50M (~49.7M)
Dataset: ~36B tokens
Context length: up to 2048 tokens
Key benchmarks: ARC-Challenge (acc 0.2116), Hellaswag (acc 0.3021), PIQA (acc 0.6496)
Discuss Telegram 1 30

Will it run?

Runs in BF16

Fits without quantisation.

  • BF16weights 0.1 GB · ready files0.6 GB
  • FP8 / INT8recommendedweights 0.0 GB0.6 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.

    # glossia/Rambley-V3-50M
    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 glossia/Rambley-V3-50M --local-dir ~/models/Rambley-V3-50M
    # The model is in ~/models. How to run it is in the model card (link above).
    Add more LoRAs to Rambley-V3-50M
  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 Rambley-V3-50M (quants, finetunes, LoRAs) will show up under the bell.

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

About the model

glossia/Rambley-V3-50M
Author
glossia
Category
Other
Released
Posted

On NeyroLearn

Hugging Face

301

Ask AI about this model

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

Comments