Set default eval batch size to 2 for LLM fine-tuning

[arnavgarg1]

Our current batch size tuning logic for eval batch size is more geared towards ECD than LLM.

For LLMs, we want to create synthetic data and push the largest possible "most computationally expensive" batches through during batch size tuning, but right now we just use batches that exist in the dataset. More on this in the future.

For now, let's set the default eval_batch_size is 2.

[github-actions]

Unit Test Results

  6 files  ±0    6 suites  ±0   1h 17m 40s ⏱️ - 2m 6s
34 tests ±0  29 ✔️ ±0    5 💤 ±0  0 ❌ ±0 
88 runs  ±0  72 ✔️ ±0  16 💤 ±0  0 ❌ ±0 

Results for commit 221fa0e. ± Comparison against base commit 6178b48.