Prime IntellectPrime RL research models
Qwen3-4B-Instruct-AlphabetSort-RL
Prime publishes this trained checkpoint as part of its alphabet-sort reinforcement-learning example.
Research release
What the release provides
The released weight files are available from Prime's official organization. The associated training recipe explains the task and model lineage.Qwen3-4B-Instruct-AlphabetSort-RL official model card
How to use it well
Practical guidance. For real record sorting, use deterministic sorting code. This model is useful for studying how a language model learns state changes and follows feedback across turns.
The task requires more than a single sorted list
The example starts from Qwen3-4B-Instruct-2507 and uses LoRA reinforcement learning without supervised warmup. The model must sort by first or last name, keep a cumulative list, and mark newly added names.Prime RL alphabet-sort training example
Specifications & access
| Detail | Value and conditions |
|---|---|
| Status | Research release |
| Access | Downloadable |
| Base model | Qwen3-4B-Instruct-2507Prime RL alphabet-sort training example |