Safetensors

Myx: Instruction Backtranslation (Reverse Model)

This Myx model is basically made according to https://arxiv.org/pdf/2308.06259.pdf

Myx is a reverse instruction generation model fine-tuned from Qwen. It is designed to generate a plausible instruction (prompt) given an output text (response). This model is a core component of the Self-Alignment with Instruction Backtranslation pipeline, enabling automatic labeling of web text for instruction tuning.

Model Description

  • Task: Given a response text, generate the corresponding instruction that would elicit such a response.
  • Base Model: Qwen3-1.7B

Usage

    # 构造输入:使用与训练时完全相同的格式
    # 训练时的格式是: response + "<|im_start|>assistant\n"
    input_text = resp + "<|im_start|>assistant\n"
    inputs = tokenizer(input_text, return_tensors="pt", truncation=True, max_length=512).to("cuda:0")
Downloads last month

-

Downloads are not tracked for this model. How to track
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support

Dataset used to train hahah1126/qwen3_1.7B_backward_v2

Paper for hahah1126/qwen3_1.7B_backward_v2