kawine commited on
Commit
6cd6ffb
·
1 Parent(s): 3ad63d6

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +8 -8
README.md CHANGED
@@ -17,11 +17,11 @@ tags:
17
  - evaluation
18
  ---
19
 
20
- # 💨🚢 SteamSHP
21
 
22
  <!-- Provide a quick summary of what the model is/does. -->
23
 
24
- SteamSHP is a preference model trained to predict human preferences, given some context and two possible responses.
25
  It can be used for NLG evaluation or to train a smaller reward model for RLHF.
26
 
27
  It is a FLAN-T5-xl model (3B parameters) finetuned on:
@@ -38,8 +38,8 @@ Here's how to use the model:
38
  >> from transformers import T5ForConditionalGeneration, T5Tokenizer
39
  >> device = 'cuda'
40
 
41
- >> tokenizer = T5Tokenizer.from_pretrained('stanfordnlp/SteamSHP-preference-model')
42
- >> model = T5ForConditionalGeneration.from_pretrained('stanfordnlp/SteamSHP-preference-model').to(device)
43
 
44
  >> input_text = "POST: Instacart gave me 50 pounds of limes instead of 5 pounds... what the hell do I do with 50 pounds of limes? I've already donated a bunch and gave a bunch away. I'm planning on making a bunch of lime-themed cocktails, but... jeez. Ceviche? \n\n RESPONSE A: Lime juice, and zest, then freeze in small quantities.\n\n RESPONSE B: Lime marmalade lol\n\n Which response is better? RESPONSE"
45
  >> x = tokenizer([input_text], return_tensors='pt').input_ids.to(device)
@@ -60,7 +60,7 @@ RESPONSE B: { second possible continuation }
60
  Which response is better? RESPONSE
61
  ```
62
 
63
- The output generated by SteamSHP will either be `A` or `B`.
64
 
65
  If the input exceeds the 512 token limit, you can use [pybsd](https://github.com/nipunsadvilkar/pySBD) to break the input up into sentences and only include what fits into 512 tokens.
66
  When trying to cram an example into 512 tokens, we recommend truncating the context as much as possible and leaving the responses as untouched as possible.
@@ -68,7 +68,7 @@ When trying to cram an example into 512 tokens, we recommend truncating the cont
68
 
69
  ## Training and Evaluation
70
 
71
- SteamSHP was only finetuned on 125K of the 392K training examples that were available, since we found that:
72
  1. When the total input length exceeded the limit (512 tokens), the loss would not converge.
73
  When possible, we crammed an example to fit under 500 tokens by truncating the context as much as possible, though some examples would still not fit despite this.
74
  We used 500 as the limit instead of 512 to allow for slight modifications to the structure of the input without any examples exceeding the actual 512 limit.
@@ -77,7 +77,7 @@ SteamSHP was only finetuned on 125K of the 392K training examples that were avai
77
  We did no such subsampling for the HH-RLHF training data.
78
 
79
  We evaluated the model on the SHP and HH-RLHF test data using accuracy, but only on the data that could be truncated to fit within 500 tokens (a total of 18621 out of 20753 available test examples).
80
- SteamSHP gets an average 72.8% accuracy across all domains:
81
 
82
  | Domain | Accuracy |
83
  | ------ | -------- |
@@ -106,7 +106,7 @@ SteamSHP gets an average 72.8% accuracy across all domains:
106
 
107
  ### Biases and Limitations
108
 
109
- Biases in the datasets used to train SteamSHP may be propagated downstream to the model predictions.
110
  Although SHP filtered out posts with NSFW (over 18) content, chose subreddits that were well-moderated and had policies against harassment and bigotry, some of the data may contain discriminatory or harmful language.
111
  Reddit users on the subreddits covered by SHP are also not representative of the broader population. They are disproportionately from developed, Western, and English-speaking countries.
112
 
 
17
  - evaluation
18
  ---
19
 
20
+ # 💨🚢 SteamSHP-XL
21
 
22
  <!-- Provide a quick summary of what the model is/does. -->
23
 
24
+ SteamSHP-XL is a preference model trained to predict human preferences, given some context and two possible responses.
25
  It can be used for NLG evaluation or to train a smaller reward model for RLHF.
26
 
27
  It is a FLAN-T5-xl model (3B parameters) finetuned on:
 
38
  >> from transformers import T5ForConditionalGeneration, T5Tokenizer
39
  >> device = 'cuda'
40
 
41
+ >> tokenizer = T5Tokenizer.from_pretrained('stanfordnlp/SteamSHP-flan-t5-xl')
42
+ >> model = T5ForConditionalGeneration.from_pretrained('stanfordnlp/SteamSHP-flan-t5-xl').to(device)
43
 
44
  >> input_text = "POST: Instacart gave me 50 pounds of limes instead of 5 pounds... what the hell do I do with 50 pounds of limes? I've already donated a bunch and gave a bunch away. I'm planning on making a bunch of lime-themed cocktails, but... jeez. Ceviche? \n\n RESPONSE A: Lime juice, and zest, then freeze in small quantities.\n\n RESPONSE B: Lime marmalade lol\n\n Which response is better? RESPONSE"
45
  >> x = tokenizer([input_text], return_tensors='pt').input_ids.to(device)
 
60
  Which response is better? RESPONSE
61
  ```
62
 
63
+ The output generated by SteamSHP-XL will either be `A` or `B`.
64
 
65
  If the input exceeds the 512 token limit, you can use [pybsd](https://github.com/nipunsadvilkar/pySBD) to break the input up into sentences and only include what fits into 512 tokens.
66
  When trying to cram an example into 512 tokens, we recommend truncating the context as much as possible and leaving the responses as untouched as possible.
 
68
 
69
  ## Training and Evaluation
70
 
71
+ SteamSHP-XL was only finetuned on 125K of the 392K training examples that were available, since we found that:
72
  1. When the total input length exceeded the limit (512 tokens), the loss would not converge.
73
  When possible, we crammed an example to fit under 500 tokens by truncating the context as much as possible, though some examples would still not fit despite this.
74
  We used 500 as the limit instead of 512 to allow for slight modifications to the structure of the input without any examples exceeding the actual 512 limit.
 
77
  We did no such subsampling for the HH-RLHF training data.
78
 
79
  We evaluated the model on the SHP and HH-RLHF test data using accuracy, but only on the data that could be truncated to fit within 500 tokens (a total of 18621 out of 20753 available test examples).
80
+ SteamSHP-XL gets an average 72.8% accuracy across all domains:
81
 
82
  | Domain | Accuracy |
83
  | ------ | -------- |
 
106
 
107
  ### Biases and Limitations
108
 
109
+ Biases in the datasets used to train SteamSHP-XL may be propagated downstream to the model predictions.
110
  Although SHP filtered out posts with NSFW (over 18) content, chose subreddits that were well-moderated and had policies against harassment and bigotry, some of the data may contain discriminatory or harmful language.
111
  Reddit users on the subreddits covered by SHP are also not representative of the broader population. They are disproportionately from developed, Western, and English-speaking countries.
112