PEFT
Safetensors
lora
retrosynthesis
chemistry
agent
retrace

ReTRACE-27B-latest

Rolling "latest" LoRA adapter of the ReTRACE retrosynthesis agent on top of Qwen/Qwen3.8-27B. Each update is pinned by a git tag; main always points to the latest.

Paper: arXiv:2610.24744 Β· Code: github.com/jxtse/retrace Β· Training data: jxtse/ReTRACE-train-data

Current version: s3-v1

  • Lineage: Qwen3.8-27B β†’ S1 v2 (chemical-judgment SFT, 819 rows) β†’ S2 v2 (agentic SFT on 167 teacher trajectories, medium reasoning) β†’ S3 (RFT-style training on the student's own harness rollouts selected by route audit; 7 optimizer steps).
  • LoRA r=8, alpha=16, dropout 0.05; attention, linear-attention and MLP projections of the language model.
  • adapter sha256 e69aa10429dfd7aaa83b3a51201832e7380cd460f720eda52b9a169e62f9b146

Evaluation (test40, medium reasoning, ReTRACE harness; LLM-judge chemistry audit, not expert ground truth)

Model native success Audit@1 strict Audit@1 lenient
Qwen3.8-27B base 35/40 4/40 11/40
ReTRACE-27B s3-v1 40/40 4/40 19/40

Small test set (40 targets); the lenient difference is not statistically significant (exact McNemar p=0.077).

Usage

Intended for use inside the ReTRACE agent harness (tool-calling retrosynthesis planning). Load with PEFT on top of Qwen/Qwen3.8-27B, or serve with vLLM --enable-lora.

Status and limitations

  • Outputs are planning suggestions; every route must be checked by a chemist before any laboratory use.

Citation

@misc{xie2026retrace,
  title         = {ReTRACE: Feedback-Guided LLM Agent in a Chemistry Harness for Retrosynthesis Planning},
  author        = {Jinxiang Xie and Yijian Zhang and Qingyuan Yi and Jianyi Wang and Hao Dong},
  year          = {2026},
  eprint        = {2610.24744},
  archivePrefix = {arXiv},
  primaryClass  = {cs.AI},
  url           = {https://arxiv.org/abs/2610.24744}
}
Downloads last month
14
Inference Providers NEW
This model isn't deployed by any Inference Provider. πŸ™‹ Ask for provider support

Model tree for jxtse/ReTRACE-27B-latest

Base model

Qwen/Qwen3.8-27B
Adapter
(362)
this model

Dataset used to train jxtse/ReTRACE-27B-latest

Paper for jxtse/ReTRACE-27B-latest