Files

25 lines
419 B
Markdown
Raw Permalink Normal View History

---
language:
- en
library_name: transformers
tags:
- qwen
- sft
- ifeval
- retaining-by-doing
---
# Qwen2.5-0.5B Instruct IFEval Half-Epoch SFT
This checkpoint was trained from `Qwen2.5-0.5B-Instruct` on `IFEvalSFTDataset`.
Training setup:
- `num_train_datapoints=4064`
- `num_epochs=1`
- effective half-epoch setting
- learning rate `1e-4`
- total batch size `64`
Observed local IFEval accuracy:
- `0.4209445585`