---
pipeline_tag: reinforcement-learning
tags:
- In-Context RL
license: mit
---

# Mixture-of-Experts Meets In-Context Reinforcement Learning

## Sources
- **Repository:** [Github](https://github.com/NJU-RL/T2MIR)
- **Paper:** [Mixture-of-Experts Meets In-Context Reinforcement Learning](https://huggingface.co/papers/2506.05426)

## Model Description
Checkpoints of T2MIR-AD and T2MIR-DPT on Cheetah-Vel using mixed datasets.  
To use the provided checkpoints, you will need to rename the file to a specific step (e.g., `"89000.pt"`) and place it into the appropriate directory (e.g., `runs/HalfCheetahVel-v0/mixed/horizon-4/exp_0-seed_3407/checkpoints/`).

## Usage

The provided code implements the T2MIR framework. See the [GitHub repository](https://github.com/NJU-RL/T2MIR) for training and evaluation instructions. Examples are provided below for training and evaluating `T2MIR-AD` and `T2MIR-DPT` on the Cheetah-Vel environment.

**Training**

To train `T2MIR-AD`, use:

```bash
cd T2MIR-AD
python train.py cheetah-vel-v0 --exp exp_0 --seed 3407
```

To train `T2MIR-DPT`, use:

```bash
cd T2MIR-DPT
python train.py cheetah-vel-v0 --exp exp_0 --seed 3407
```

**Evaluation**

To evaluate `T2MIR-AD`, use:

```bash
cd T2MIR-AD
python eval.py cheetah-vel-v0 1 --exp exp_0 --seed 3407 --start-ckpt 89000 --stop-ckpt 89000 --seed-eval 10
```

To evaluate `T2MIR-DPT`, use:

```bash
cd T2MIR-DPT
python eval_online.py cheetah-vel-v0 1 --exp exp_0 --seed 3407 --start-ckpt 59000 --stop-ckpt 59000 --seed-eval 10
```