Skip to content
Language models

Pascal-15M-v1.0

a micro-transformer for numerical sequence continuation

2 min readupdated

Pascal-15M-v1.0 from AtomixLabs is a compact 17.7M-parameter language model that can't write prose or code, but excels at spotting patterns in numbers and continuing sequences. Under the hood, it uses a standard 12-layer LLaMA architecture trained on roughly 364M tokens of numerical data. Unlike standard LLMs that glue digits into arbitrary chunks via BPE tokenization, it uses an exact 32-token vocabulary—one token per digit, separator, and sign. Researchers and developers can use it as a lightweight base for fine-tuning on specific numerical tasks.

Key features

  • 32-token vocabulary: includes only digits 0–9, comma, minus sign, period, slash (/), and special tokens, eliminating rounding errors and tokenization artifacts.
  • Handles diverse sequence types: uncovers rules in arithmetic and geometric progressions, fractional approximations (such as Fibonacci numbers), alternating streams, and oscillators.
  • Native LLaMA architecture: runs out of the box in PyTorch and Transformers with no custom CUDA kernels or trust_remote_code=True required.
  • Controlled output formatting: the tiny vocabulary makes it easy to constrain generation via logit masking, blocking invalid syntax like /0 or duplicate signs.

What you can do

  • Telemetry and sensors: fine-tune the checkpoint on sensor readings or monitoring logs to forecast upcoming values.
  • Financial ticks: adapt the model to analyze discrete price levels, spreads, and rate changes.
  • Dynamic system simulation: use it as a foundation for kinematics, coupled oscillations, or empirical symbolic regression.

Example

from transformers import AutoModelForCausalLM, AutoTokenizer

model = AutoModelForCausalLM.from_pretrained("AtomixLabs/Pascal-15M-v1.0", device_map="auto")
tokenizer = AutoTokenizer.from_pretrained("AtomixLabs/Pascal-15M-v1.0")
inputs = tokenizer("1, 1/2, 2/3, 3/5, 5/8,", return_tensors="pt").to(model.device)
sep_id = tokenizer.convert_tokens_to_ids(",")
out = model.generate(**inputs, max_new_tokens=24, do_sample=False, eos_token_id=sep_id)
print(tokenizer.decode(out[0], skip_special_tokens=True))  # 1,1/2,2/3,3/5,5/8,8/13
Read moreDetails
Parameters: 17,716,608 (hidden_size 384, 12 layers, 12 attention heads, 512-token context).
Note: the model wasn't trained with an token, so stopping must be handled manually via commas or max_new_tokens.
Limitations: doesn't understand natural text, is sensitive to whitespace and formatting, and tends to hallucinate spurious patterns when fed random noise.
License: Apache-2.0.
Discuss Telegram 1 55

Will it run?

Runs in F32

Fits without quantisation.

  • F32weights 0.1 GB · ready files0.8 GB
  • BF16weights 0.0 GB0.8 GB
  • FP8 / INT8recommendedweights 0.0 GB0.8 GB
  • 4-bitweights 0.0 GB0.7 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.

Language models: with a 8192-token context.

Estimate: weights plus working memory. Long videos, high resolutions and long contexts need more.

Run it on Massed Compute

A ready script for the machine we advise above: paste it and the model downloads and starts.

  1. 1. Rent A30

    $0.26 an hour with the coupon. Pick the Ubuntu image when you create the machine.

    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.

    # AtomixLabs/Pascal-15M-v1.0 — A30, FP8 / INT8
    # NVIDIA driver: CUDA 13 needs 580+. If older or missing, the current nvidia-open is installed (keep yours: KEEP_DRIVER=1)
    DRV=$(nvidia-smi --query-gpu=driver_version --format=csv,noheader 2>/dev/null | head -1 | cut -d. -f1)
    case "$DRV" in ""|*[!0-9]*) DRV=0 ;; esac
    if [ "$DRV" -lt 580 ] && [ -z "${KEEP_DRIVER:-}" ]; then
      . /etc/os-release && \
      curl -fsSLO "https://developer.download.nvidia.com/compute/cuda/repos/ubuntu${VERSION_ID//./}/x86_64/cuda-keyring_1.1-1_all.deb" && \
      sudo dpkg -i cuda-keyring_1.1-1_all.deb && rm -f cuda-keyring_1.1-1_all.deb && \
      sudo apt-get update && \
      sudo DEBIAN_FRONTEND=noninteractive apt-get install -y nvidia-open || \
      { echo "Could not install the NVIDIA driver: see the error above. To keep the current one, run with KEEP_DRIVER=1."; exit 1; }
      echo "NVIDIA driver updated. Reboot the machine (sudo reboot) and run this same script again."
      exit 0
    fi
    curl -LsSf https://astral.sh/uv/install.sh | sh
    source "$HOME/.local/bin/env"
    uv venv ~/vllm --python 3.12 && source ~/vllm/bin/activate
    uv pip install vllm
    vllm serve AtomixLabs/Pascal-15M-v1.0 --max-model-len 8192 --quantization fp8
    Add more LoRAs to Pascal-15M-v1.0
  3. 3. Use it

    The server listens on port 8000 and speaks the OpenAI API: http://localhost:8000/v1 on the machine. From your computer: ssh -L 8000:localhost:8000 user@machine-IP.

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.

  • MTP (multi-token prediction)0.34.0from November 16, 2026new release

    1.5–2× generation speed

    If the model has MTP layers (DeepSeek-V3/R1, Qwen3-Next, GLM, MiMo and others), it predicts several tokens per step itself.

    How to enable: vLLM / SGLang: --speculative-config '{"method": "mtp", "num_speculative_tokens": 1}'. Check the model card.

    Earlier generations (2)
    • EAGLE-3 — Speculative decoding with a small draft head. For models without their own MTP layers.
    • EAGLE-2 / Medusa — Earlier draft-head generations. Still work if that is all you find.
  • FP8 KV cache0.34.0from November 16, 2026new release

    2× more context in the same memory

    Keeps the attention cache in 8 bits, so long contexts fit where they did not.

    How to enable: vLLM: --kv-cache-dtype fp8; llama.cpp: -ctk q8_0 -ctv q8_0.

  • vLLM / SGLang0.34.0from November 16, 2026new release

    many times more tokens/s on a server

    Server engines: batching, FlashAttention, FP8 and speculative decoding out of the box.

    How to enable: vllm serve <model> — an OpenAI-compatible API.

    Earlier generations (1)
    • llama.cpp — GGUF models on any GPU, Mac or even CPU; some layers can stay in RAM.

Model family

New posts about versions of Pascal-15M-v1.0 (quants, finetunes, LoRAs) will show up under the bell.

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

About the model

AtomixLabs/Pascal-15M-v1.0
Author
AtomixLabs
Category
Language models
Released
Posted

On NeyroLearn

Hugging Face

551

Ask AI about this model

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

Comments