metadata
license: apache-2.0
datasets:
- allenai/dolma
- allenai/tulu-v2-sft-mixture-olmo-4096
- argilla/ultrafeedback-binarized-preferences-cleaned
language:
- en
OLMo-1B-0724 Instruct
This is a version of OLMo-1B-0724-hf that has undergone SFT and DPO training. See the SFT model card for details on SFT training.
This model is initialised from OLMo-1B-0724-SFT-hf, and then DPO trained on a cleaned ultrafeedback dataset for 3 epochs with a batch size of 32, beta of 0.1, linear warmup for 10% of training, and then linear cooldown.
Evals are as follows:
| Metric | OLMo-1B-0724-hf | OLMo-1B-0724-SFT-hf | OLMo-1B-0724-Instruct-hf (this model!) |
|---|---|---|---|
| MMLU 0-shot | 25.0 | 36.0 | 36.7 |
| GSM8k CoT 8-shot | 7.0 | 12.5 | 12.5 |
| BBH CoT 3-shot | 22.5 | 27.2 | 30.6 |
| HumanEval P@10 | 16.0 | 21.2 | 22.0 |
| AlpacaEval 1 | - | 41.5 | 50.9 |
| AlpacaEval 2 LC | - | 2.7 | 2.5 |
| Toxigen % Toxic | 80.3 | 59.7 | 14.1 |
| TruthfulQA %Info+True | 23.0 | 40.9 | 42.2 |
| IFEval Loose Acc | 20.5 | 26.1 | 24.2 |
| XSTest F1 | 67.6 | 81.9 | 79.8 |
| Average of above metrics | 25.2 | 33.0 | 38.7 |
Model training and evaluation was performed using Open-instruct, so check that out for more details on evaluation.