RL training smoke tests

[winglian]

⚠️ Please check that this feature request hasn't been suggested before.

• [X] I searched previous Ideas in Discussions didn't find any similar feature requests.
• [X] I searched previous Issues didn't find any similar feature requests.

🔖 Feature description

We need to add smoke tests and integration tests for axolotl w. trl to ensure we have compatibility with them as both projects iterate. We can start with the supported methods such as DPO, IPO, KTO-pairs. Let's just use some tiny models like tiny-llama or sheared llama.

✔️ Solution

add tests under e2e for dpo, kto_pair, and iso

❓ Alternatives

No response

📝 Additional Context

No response

Acknowledgements

• [X] My issue title is concise, descriptive, and in title casing.
• [X] I have searched the existing issues to make sure this feature has not been requested yet.
• [X] I have provided enough information for the maintainers to understand and evaluate this request.

[JohanWork]

Was looking into to trying to solve this issue, but believe it already is solved here https://github.com/OpenAccess-AI-Collective/axolotl/blob/main/tests/e2e/test_dpo.py or @winglian