Skip to content
Other

ReTRACE-27B-latest

an agentic LoRA adapter for chemical synthesis planning

2 min read

This is a continuously updated LoRA (a lightweight add-on for a base model) adapter built on top of the Qwen/Qwen3.8-27B language model for agentic retrosynthesis planning (finding synthetic routes for chemical compounds). It is trained to handle chemical reasoning and construct reaction sequences using tool-calling. The s3-v1 release uses a multi-stage training pipeline that combines SFT (supervised fine-tuning) and RFT (reinforcement fine-tuning on the agent's own trajectories). It is built for researchers, chemists, and developers working on lab planning automation.

Key features

  • Built on the Qwen3.8-27B base model using the PEFT library.
  • Trained across multiple stages, from chemistry SFT to RFT on self-generated rollouts.
  • Adapter configuration: LoRA rank r=8, alpha=16, and dropout 0.05 across attention, linear-attention, and MLP projections.
  • Tailored specifically for tool-calling workflows in the ReTRACE agentic environment.

What you can do

  • Load the adapter on top of the Qwen3.8-27B base model via the PEFT library for local experiments.
  • Serve the model with adapter support using vLLM and the --enable-lora flag.
  • Integrate the model into the ReTRACE agent harness to automate chemical synthesis route planning.

Example

Run the model inside the ReTRACE agent harness to generate retrosynthesis plans. Keep in mind that all outputs are preliminary plans and must be verified by a professional chemist before any lab execution.

Read moreDetails
Metrics (test40): native success rate 40/40 (vs. 35/40 for the base model), Audit@1 lenient 19/40 (vs. 11/40)
Requirements: compatibility with Qwen/Qwen3.8-27B and the peft library
Limitations: outputs require mandatory review by an expert chemist before practical use
Discuss Telegram 0

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 files58 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.

This is a LoRA: counted together with its base model Qwen3.8-27B.

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.

    # jxtse/ReTRACE-27B-latest
    # ===== Tokens: paste yours between the quotes =====
    export HF_TOKEN=""        # huggingface.co/settings/tokens, for gated models
    # ==========================================
    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 Qwen/Qwen3.8-27B --local-dir ~/models/Qwen3.8-27B
    hf download jxtse/ReTRACE-27B-latest --local-dir ~/models/ReTRACE-27B-latest
    # The model is in ~/models. How to run it is in the model card (link above).
    Add more LoRAs to Qwen3.8-27B
  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

Requires the base model

Qwen/Qwen3.8-27B

This model is a LoRA / adapter. Download the base model first, then load this one on top.

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

jxtse/ReTRACE-27B-latest
Author
jxtse
Category
Other
Released
Posted

On NeyroLearn

Hugging Face

00

Ask AI about this model

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

Comments

Robotics

Jevol-2B

Jevol-2B