Merge branch 'main' of https://huggingface.co/Leogrin/eleuther-pythia1b-hh-sft into main
Browse files
README.md
ADDED
|
@@ -0,0 +1,54 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
language:
|
| 3 |
+
- en
|
| 4 |
+
tags:
|
| 5 |
+
- pytorch
|
| 6 |
+
- causal-lm
|
| 7 |
+
- pythia
|
| 8 |
+
license: apache-2.0
|
| 9 |
+
datasets:
|
| 10 |
+
- Anthropic/hh-rlhf
|
| 11 |
+
---
|
| 12 |
+
|
| 13 |
+
# Infos
|
| 14 |
+
|
| 15 |
+
Pythia-1b supervised finetuned with Anthropic-hh-rlhf dataset for 1 epoch.
|
| 16 |
+
|
| 17 |
+
[wandb log](https://wandb.ai/pythia_dpo/Pythia_DPO_new/runs/xk2ub7ig?workspace=user-leogrin)
|
| 18 |
+
|
| 19 |
+
See [Pythia-1b](https://huggingface.co/EleutherAI/pythia-1b) for model details [(paper)](https://arxiv.org/abs/2101.00027).
|
| 20 |
+
|
| 21 |
+
|
| 22 |
+
# Benchmark raw results:
|
| 23 |
+
|
| 24 |
+
Results for the base model are taken from the [Pythia paper](https://arxiv.org/abs/2101.00027).
|
| 25 |
+
|
| 26 |
+
## Zero shot
|
| 27 |
+
|
| 28 |
+
|
| 29 |
+
| Task | 1B_base | 1B_sft |
|
| 30 |
+
|------------------|----------------|----------------|
|
| 31 |
+
| Lambada (OpenAI) | 0.562 ± 0.007 | 0.563 ± 0.007 |
|
| 32 |
+
| PIQA | 0.707 ± 0.011 | 0.711 ± 0.011 |
|
| 33 |
+
| WinoGrande | 0.537 ± 0.014 | 0.534 ± 0.014 |
|
| 34 |
+
| WSC | 0.365 ± 0.047 | 0.365 ± 0.047 |
|
| 35 |
+
| ARC - Easy | 0.569 ± 0.010 | 0.583 ± 0.010 |
|
| 36 |
+
| ARC - Challenge | 0.244 ± 0.013 | 0.248 ± 0.013 |
|
| 37 |
+
| SciQ | 0.840 ± 0.012 | 0.847 ± 0.011 |
|
| 38 |
+
| LogiQA | 0.223 ± 0.016 | -- |
|
| 39 |
+
|
| 40 |
+
## Five shot
|
| 41 |
+
|
| 42 |
+
| Task | 1B_base | 1B_sft |
|
| 43 |
+
|------------------|----------------|----------------|
|
| 44 |
+
| Lambada (OpenAI) | 0.507 ± 0.007 | 0.4722 ± 0.007 |
|
| 45 |
+
| PIQA | 0.705 ± 0.011 | 0.7165 ± 0.0105|
|
| 46 |
+
| WinoGrande | 0.532 ± 0.014 | 0.5343 ± 0.014 |
|
| 47 |
+
| WSC | 0.365 ± 0.047 | 0.5000 ± 0.0493|
|
| 48 |
+
| ARC - Easy | 0.594 ± 0.010 | 0.6010 ± 0.010 |
|
| 49 |
+
| ARC - Challenge | 0.259 ± 0.013 | 0.2679 ± 0.0129|
|
| 50 |
+
| SciQ | 0.920 ± 0.009 | 0.9100 ± 0.0091|
|
| 51 |
+
| LogiQA | 0.227 ± 0.016 | N/A |
|
| 52 |
+
|
| 53 |
+
|
| 54 |
+
|