Prime IntellectPrime RL research models
Qwen3.5-0.8B-Reverse-Text-SFT
This small supervised fine-tune uses Prime's Reverse-Text-SFT dataset. Prime made it as a starting policy for internal training-framework checks and says it is not for general use.
Research release
What the release provides
The recipe runs ten supervised steps on one H200. It retains the unchanged Qwen3.5 chat template with thinking disabled. The vision tower remains frozen and unchanged in the released weights.Qwen3.5-0.8B-Reverse-Text-SFT official model card
How to use it well
Practical guidance. Use this model to examine a training pipeline at small scale. Its purpose is to leave room for later learning, so low initial task performance is expected.
Specifications & access
| Detail | Value and conditions |
|---|---|
| Status | Research release |
| Access | Downloadable |
| Repository license | Apache 2.0Qwen3.5-0.8B-Reverse-Text-SFT official model card |
| Base model | Qwen3.5-0.8BQwen3.5-0.8B-Reverse-Text-SFT official model card |
Related models
- Qwen3.5-0.8B-Reverse-Text-RL
Prime Intellect ยท Prime RL research models
A short reinforcement-learning run for Qwen3.5 training-system checks.
Launch date not confirmedResearch releaseDownloadable