Skip to content
Other

Transformer-134M-BabiStories

A Baseline Model for Architecture Comparisons

2 min readupdated

Built to study and benchmark architectures for the Memory Mosaics v2 project, this compact research model was trained from scratch on the BabiStories children's stories dataset. Rather than relying on pretrained weights from large models like Llama or Mistral, it builds from the ground up while using the Mistral tokenizer. With 134 million parameters, it serves as a standard Transformer baseline to evaluate the new Memory Mosaics architecture, targeting NLP researchers who study the efficiency of different architectural approaches.

Key features

  • Architecture: 12 layers with a width of 768, using RMSNorm, causal self-attention with RoPE, and SwiGLU FFN.
  • Trained from scratch: 15,502 steps (127 million tokens total) using the AdamW optimizer in bf16 precision.
  • Metrics: Achieves a final held-out validation loss of 1.7824 for direct comparison against its paired Memory Mosaics v2 model.
  • Efficiency: Trained in 21 minutes on a single RTX 5090 GPU at roughly 102,883 tokens per second.

What you can do

  • Run text generation: Use the loading script to test prompt continuations and generate stories.
  • Compare architectures: Benchmark training speed, VRAM usage, and loss metrics directly against Memory Mosaics v2.
  • Study small LLMs: Explore how compact language models behave when trained on specialized datasets.

Example

Run text generation using the script from the repository:

git clone [https://github.com/facebookresearch/MemoryMosaics](https://github.com/facebookresearch/MemoryMosaics)
git -C MemoryMosaics checkout c2ca0e88c5c89fc18a0a9465d6f7a1c3c8ea8faf
pip install torch safetensors tokenizers pyyaml huggingface_hub
wget [https://huggingface.co/Eureka-COM/Transformer-134M-BabiStories/resolve/main/load_and_generate.py](https://huggingface.co/Eureka-COM/Transformer-134M-BabiStories/resolve/main/load_and_generate.py)
python load_and_generate.py --upstream ./MemoryMosaics --repo Eureka-COM/Transformer-134M-BabiStories --prompt "Once upon a time, a little fox"
Read moreDetails
Parameters: 134M (134,105,856)
Context: 512 tokens, 32,000 vocabulary
VRAM usage: 7.4 GB peak (PyTorch, batch 16 × 512)
License: Apache-2.0
Examples
Discuss Telegram 0 53

Will it run?

Runs in F32

Fits without quantisation.

  • F32weights 0.5 GB · ready files1.0 GB
  • BF16weights 0.2 GB0.8 GB
  • FP8 / INT8recommendedweights 0.1 GB0.6 GB
  • 4-bitweights 0.1 GB0.6 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.

    # Eureka-COM/Transformer-134M-BabiStories
    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 Eureka-COM/Transformer-134M-BabiStories --local-dir ~/models/Transformer-134M-BabiStories
    # 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 Transformer-134M-BabiStories (quants, finetunes, LoRAs) will show up under the bell.

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

About the model

Eureka-COM/Transformer-134M-BabiStories
Author
Eureka-COM
Category
Other
Released
Posted

On NeyroLearn

Hugging Face

530

Ask AI about this model

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

Comments