Skip to content
Computer vision

DaisyRose-27B

a multimodal reasoning model built on Qwen3.8-27B

2 min readupdated

RadixArk's DaisyRose-27B is a multimodal model for complex agentic tasks and reasoning, trained exclusively with RL (reinforcement learning), skipping both supervised fine-tuning and distillation. It retains the architecture, tokenizer, and 256K context window of the base Qwen3.8-27B model, supporting both text and image inputs. Released under the Apache-2.0 license in BF16, FP8, and NVFP4 formats, it serves as a drop-in replacement for the base version in existing inference pipelines.

Key features

  • Pure RL training: Trained on 1,200 curated long-horizon agentic tasks with sessions up to 1M tokens.
  • Multimodality: Handles both image and text understanding with a 262,144-token context length.
  • Benchmark gains: Terminal Bench 4.1 score jumps from 5.6 to 12.0 compared to the base model.
  • Flexible deployment: Available in BF16, FP8, and NVFP4 weights to fit different hardware setups.

What you can do

  • Tackle complex tasks: Power autonomous agents for coding, code analysis, and engineering workflows.
  • Analyze multimodal data: Process long texts and images across multi-step scenarios.
  • Deploy with SGLang: Run the model locally or on a server using an optimized backend with DSpark speculative decoding.

Example

Sample request to a local server via the openai client to solve a logic puzzle with chain-of-thought output:

from openai import OpenAI

client = OpenAI(base_url="http://localhost:30000/v1", api_key="none")
SAMPLING = dict(temperature=1.0, top_p=0.95, extra_body={"top_k": 20, "min_p": 0.0})

resp = client.chat.completions.create(
    model="DaisyRose-27B",
    messages=[{"role": "user", "content": "How many positive divisors does 360 have?"}],
    **SAMPLING,
)
print(resp.choices[0].message.reasoning_content)  # ходы рассуждений
print(resp.choices[0].message.content)            # финальный ответ
Read moreDetails
• Architecture: based on Qwen3.8-27B (64 decoder layers, hybrid linear-attention and full-attention, multimodal encoder).
• Parameters: 27.8B total parameters, including a 0.46B encoder and a 0.42B MTP head.
• Training: 18 days on 144 H200 GPUs using the Miles framework.
• Memory requirements: weights take roughly 52 GB in BF16, 29 GB in FP8, and 23 GB in NVFP4.
Discuss Telegram 277 918

Will it run?

Runs in 4-bit

Full precision does not fit, a quantised version does: quality barely changes down to 8 bits, more noticeably at 4 bits.

  • BF16weights 52 GB · ready files57 GB
  • FP8 / INT8recommendedweights 26 GB29 GB
  • 4-bitweights 15 GB17 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.

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.

    # RadixArk/DaisyRose-27B
    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 RadixArk/DaisyRose-27B --local-dir ~/models/DaisyRose-27B
    # The model is in ~/models. How to run it is in the model card (link above).
    Add more LoRAs to DaisyRose-27B
  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 Qwen3.8-27B (quants, finetunes, LoRAs) will show up under the bell.

Quantised versions 12All on Hugging Face
Fine-tunes 12All on Hugging Face
LoRAs and adapters 2All on Hugging Face

About the model

RadixArk/DaisyRose-27B
Author
RadixArk
Category
Computer vision
Released
Posted

On NeyroLearn

Hugging Face

918277

Ask AI about this model

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

Comments