Skip to content
Language models

PowerRL R1-Distill 1.5B

a math model with adjustable reasoning depth

2 min readupdated

A compact 1.78B-parameter language model built for Olympiad-level math problems with configurable reasoning effort. Based on DeepSeek-R1-Distill-Qwen-1.5B, it was fine-tuned by steviel using reinforcement learning (RL) on the DAPO-Math-17k dataset. Its standout feature is letting you pick one of five effort levels via the system prompt (from minimal Low up to maximum Xhigh), giving you direct control over the trade-off between accuracy and token usage. A solid fit for developers and researchers who need tight control over token budgets when running lightweight reasoning models locally.

Key features

  • 5 reasoning levels set via the system prompt: Low, Limited, Mid, High, and Xhigh (four adapted from the ORBIT method, with Limited added specifically for this model).
  • Direct speed-vs-accuracy trade-off: at Low, the model uses an average of 723–898 tokens, while High uses around 7,100 tokens and boosts accuracy nearly 2.5x.
  • Full merged weights in bf16 format (training step 1500), with the tokenizer and chat template inherited directly from the base model.
  • Generation context up to 16,384 tokens, outputting final answers in standard \boxed{} format.

What you can do

  • Integrate it into a local Python math-solving or verification service using the transformers library, switching effort levels based on problem difficulty.
  • Run inference optimization experiments to find the lowest sufficient reasoning level (such as Limited or Mid) to save compute.

Example

from transformers import AutoModelForCausalLM, AutoTokenizer

repo = "steviel/powerrl-r1distill-1.5b-16k-step1500"
tok = AutoTokenizer.from_pretrained(repo)
model = AutoModelForCausalLM.from_pretrained(repo, torch_dtype="auto", device_map="auto")

messages = [
    {"role": "system", "content": "You have limited time to think and respond to the user’s query. ..."},
    {"role": "user", "content": "Find n <= 100 such that n^2 + 1 is divisible by 5. Let's think step by step and output the final answer within \\boxed{}."},
]
inputs = tok.apply_chat_template(messages, add_generation_prompt=True, return_tensors="pt").to(model.device)
out = model.generate(inputs, max_new_tokens=16384, do_sample=True, temperature=0.6, top_p=0.95)
print(tok.decode(out[0, inputs.shape[1]:], skip_special_tokens=True))
Read moreDetails
AIME 2024 benchmark: 43.0% accuracy at High (7,109 tokens), 35.7% at Mid (4,054 tokens), and 17.9% at Low (898 tokens).
AIME 2025 benchmark: 31.1% at High (7,073 tokens), 27.9% at Mid (3,248 tokens), and 14.9% at Low (723 tokens).
Training: 1,500 steps (128 problems × 16 samples) with an accuracy reward and a length penalty scaled to the chosen effort level.
Limitations: trained and evaluated exclusively on English-language Olympiad math; responses over 12,000 tokens are accurate only 9% of the time (compared to 89% for responses under 4,000 tokens).
Discuss Telegram 0 35

Will it run?

Runs in BF16

Fits without quantisation.

  • BF16weights 3.3 GB · ready files4.1 GB
  • FP8 / INT8recommendedweights 1.7 GB2.5 GB
  • 4-bitweights 0.9 GB1.7 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 · BF16 — 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.

    # steviel/powerrl-r1distill-1.5b-16k-step1500 — 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 steviel/powerrl-r1distill-1.5b-16k-step1500 --max-model-len 8192 --quantization fp8
  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 DeepSeek-R1-Distill-Qwen-1.5B (quants, finetunes, LoRAs) will show up under the bell.

Quantised versions 12All on Hugging Face
Fine-tunes 12All on Hugging Face
LoRAs and adapters 4All on Hugging Face

About the model

steviel/powerrl-r1distill-1.5b-16k-step1500
Author
steviel
Category
Language models
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