stellaathena commited on
Commit
3e1c92c
1 Parent(s): 5640038

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +19 -11
README.md CHANGED
@@ -48,25 +48,33 @@ GPT-Neo was trained as an autoregressive language model. This means that its cor
48
  GPT-Neo was trained on the Pile, a dataset known to contain profanity, lewd, and otherwise abrasive language. Depending on your usecase GPT-Neo may produce socially unacceptable text. See Sections 5 and 6 of the Pile paper for a more detailed analysis of the biases in the Pile.
49
 
50
  As with all language models, it is hard to predict in advance how GPT-Neo will respond to particular prompts and offensive content may occur without warning. We recommend having a human curate or filter the outputs before releasing them, both to censor undesirable content and to improve the quality of the results.
 
51
  ## Eval results
52
 
53
- ### Language Modeling Baselines
 
 
54
 
55
- EleutherAI is currently in the process of carrying out further evaluations of GPT-Neo. The following table should be considered a work-in-progress. If you would like to contribute evaluations you have done, please reach out on our Discord.
 
 
 
 
 
56
 
57
- | Model and Size | Pile BPB | Pile PPL | Wikitext PPL. |
58
- | ---------------- | ------------- | ------------- | -------------- |
59
- | GPT-Neo 1.3B | 0.7527 | 6.159 | 13.10 |
60
- | GPT-3 1.3B | ------ | ----- | ----- |
61
- | GPT-2 1.5B | 1.0468 | ----- | 17.48 |
62
- | **GPT-Neo 2.7B** | **0.7165** | **5.646** | **11.39** |
63
- | GPT-3 2.7B | 0.9631 | ----- | ----- |
64
- | GPT-3 175B | 0.7177 | ----- | ----- |
65
 
66
- All GPT-2 and GPT-3 scores are from their respective papers, except for the Pile test results which are from the Pile paper.
 
 
 
 
 
67
 
68
  ### Down-Stream Applications
69
 
 
 
70
  ### BibTeX entry and citation info
71
 
72
  ```bibtex
 
48
  GPT-Neo was trained on the Pile, a dataset known to contain profanity, lewd, and otherwise abrasive language. Depending on your usecase GPT-Neo may produce socially unacceptable text. See Sections 5 and 6 of the Pile paper for a more detailed analysis of the biases in the Pile.
49
 
50
  As with all language models, it is hard to predict in advance how GPT-Neo will respond to particular prompts and offensive content may occur without warning. We recommend having a human curate or filter the outputs before releasing them, both to censor undesirable content and to improve the quality of the results.
51
+
52
  ## Eval results
53
 
54
+ All evaluations were done using our [evaluation harness](https://github.com/EleutherAI/lm-evaluation-harness). Some results for GPT-2 and GPT-3 are inconsistent with the values reported in the respective papers. We are currently looking into why, and would greatly appreciate feedback and further testing of our eval harness. If you would like to contribute evaluations you have done, please reach out on our [Discord](https://discord.gg/vtRgjbM).
55
+
56
+ ### Linguistic Reasoning
57
 
58
+ | Model and Size | Pile BPB | Pile PPL | Wikitext PPL | Lambada PPL | Lambada Acc | Winogrande | Hellaswag |
59
+ | ---------------- | ---------- | ---------- | ------------- | ----------- | ----------- | ---------- | ----------- |
60
+ | GPT-Neo 1.3B | 0.7527 | 6.159 | 13.10 | 7.498 | 57.23% | 55.01% | 38.66% |
61
+ | GPT-2 1.5B | 1.0468 | ----- | 17.48 | 10.634 | 51.21% | 59.40% | 40.03% |
62
+ | **GPT-Neo 2.7B** | **0.7165** | **5.646** | **11.39** | **5.626** | **62.22%** | **56.50%** | **42.73%** |
63
+ | GPT-3 Ada | 0.9631 | ----- | ----- | 9.954 | 51.60% | 52.90% | 35.93% |
64
 
65
+ ### Physical and Scientific Reasoning
 
 
 
 
 
 
 
66
 
67
+ | Model and Size | MathQA | PubMedQA | Piqa |
68
+ | ---------------- | ---------- | ---------- | ----------- |
69
+ | GPT-Neo 1.3B | 24.05% | 54.40% | 71.11% |
70
+ | GPT-2 1.5B | 23.64% | 58.33% | 70.78% |
71
+ | **GPT-Neo 2.7B** | **24.72%** | **57.54%** | **72.14%** |
72
+ | GPT-3 Ada | 24.29% | 52.80% | 68.88% |
73
 
74
  ### Down-Stream Applications
75
 
76
+ TBD
77
+
78
  ### BibTeX entry and citation info
79
 
80
  ```bibtex