jiosephlee/assay-transfer-tool

TEXT GENERATIONConcurrent Unit Cost:1Model Size:8BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Jul 16, 2026License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Cold

The jiosephlee/assay-transfer-tool is an 8 billion parameter full-parameter SFT (Supervised Fine-Tuning) of the Qwen3ForCausalLM Intern-S1-mini language backbone, developed by jiosephlee. This model is specifically designed for binary assay-transfer prediction on small molecules, taking SMILES input and a paired-choice prompt to output either (A) or (B). It utilizes the Intern SMILES-aware tokenizer and is optimized for chemical informatics tasks, achieving a binary macro-F1 of 0.6612 on its validation set. Its primary use case is predicting assay outcomes for small molecules, distinguishing it from general-purpose language models.

Loading preview...

Overview

The jiosephlee/assay-transfer-tool is an 8 billion parameter language model, specifically a full-parameter Supervised Fine-Tuning (SFT) of the Qwen3ForCausalLM Intern-S1-mini backbone (jiosephlee/Intern-S1-mini-lm). Its core function is binary assay-transfer prediction for small molecules. Given a molecule's SMILES representation and a paired-choice prompt, the model predicts either (A) or (B).

Key Capabilities

  • Specialized Chemical Prediction: Designed for assay-transfer prediction in chemistry, a niche application not typically covered by general LLMs.
  • SMILES-aware Tokenization: Utilizes the Intern SMILES-aware tokenizer, crucial for accurate chemical structure processing.
  • Optimized for Binary Classification: Fine-tuned to output clear (A) or (B) responses for paired-choice prompts.
  • Performance: Achieves a binary macro-F1 score of 0.6612 and an accuracy of 0.7046 on its validation set, with a 1.00 parse rate.

What Makes This Different?

Unlike general-purpose LLMs, this model is not designed for conversational AI, code generation, or broad text understanding. Its unique differentiator is its highly specialized focus on chemical assay prediction. It leverages a robust language model backbone but is meticulously fine-tuned on a specific dataset (jiosephlee/assay-transfer-intern) to excel at a very particular task within cheminformatics. This makes it a domain-specific tool rather than a versatile generalist.

Should I Use This?

  • Use this if: Your primary need is to predict binary assay outcomes for small molecules based on SMILES input. This model is a strong candidate for tasks requiring automated chemical property prediction or drug discovery workflows where assay transfer is relevant.
  • Do not use this if: You need a general-purpose language model for tasks like text generation, summarization, question answering, or coding. This model is highly specialized and will not perform well outside its intended chemical prediction domain.