File size: 2,422 Bytes
014a75b b7d6fbe 014a75b b7d6fbe 014a75b 15d5fd5 014a75b 2f6fb76 15d5fd5 64ef5a9 014a75b |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 |
---
license: apache-2.0
datasets:
- allenai/dolma
- allenai/tulu-v2-sft-mixture-olmo-4096
language:
- en
---
# OLMo-1B-0724 SFT
[OLMo-1B-0724-hf](https://huggingface.co/allenai/OLMo-1B-0724-hf) finetuned for 5 epochs with a learning rate of 1e-5 on the Tulu 2 dataset - specifically [this version](https://huggingface.co/datasets/allenai/tulu-v2-sft-mixture-olmo-4096).
I used a batch size of 1, 128 grad accumulation steps. Linear warmup for the first 3% of training then linear decay to 0.
I've additionally [released an 'instruct' version](https://huggingface.co/hamishivi/OLMo-1B-0724-Instruct-hf) which has additionally gone through DPO training.
This model is generally more performant (see the metrics below), so check it out!
Evals are as follows:
| Metric | [OLMo-1B-0724-hf](https://huggingface.co/allenai/OLMo-1B-0724-hf) | **[OLMo-1B-0724-SFT-hf](https://huggingface.co/hamishivi/OLMo-1B-0724-SFT-hf) (this model!)** | [OLMo-1B-0724-Instruct-hf](https://huggingface.co/hamishivi/OLMo-1B-0724-Instruct-hf)|
|---------------------------|-----------------|---------------------|-------------------------|
| 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](https://github.com/allenai/open-instruct), so check that out for more details on evaluation. |