Skip to content

Q-RWKV-6 32B Instruct Preview

A 32B RWKV variant built by weight conversion

Visit official site

What Q-RWKV-6 32B Instruct Preview is

The model is described as an RWKV variant built by moving Qwen 32B Instruct weights into a custom QRWKV6 architecture.

What you can do with it

  • Run text classification and sentiment analysis
  • Translate corpora and retrieve cross language information
  • Give chatbots more natural language interaction

Model notes

  • Training covers more than 30 languages
  • Inference cost is stated to drop by over 1000 times
  • The conversion finished in eight hours

Getting started

  • Visit Hugging Face or Featherless.ai
  • Download the weights and code
  • Configure the environment and load text input

Pros & cons

✓ What we like

  • Conversion avoids training from scratch
  • The approach scales to larger Transformer based models

! What to watch out for

  • Local deployment needs configured hardware and software
  • The build used 16 AMD MI300X GPUs with 192GB VRAM each

FAQ

Who developed it?

Recursal AI is named as the developer, working with the RWKV and EleutherAI open source communities.

Where can the weights be found?

The steps name Hugging Face and Featherless.ai as places to find the model.

Last reviewed: 2026-11-15

More AI coding tools tools

View all →

How we review