Text Generation
Transformers
Safetensors
English
stablelm
conversational
Inference Endpoints
euclaise commited on
Commit
00f4710
·
verified ·
1 Parent(s): 01e78cf

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +24 -1
README.md CHANGED
@@ -116,4 +116,27 @@ As I expected, it improves GSM8K, but doesn't do much to ARC.
116
  - Epochs: 6
117
  - Learning rate: 1e-5
118
  - Learning rate schedule: One Cycle, cosine, no cycle_momentum
119
- - Regularization weight: 0.1
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
116
  - Epochs: 6
117
  - Learning rate: 1e-5
118
  - Learning rate schedule: One Cycle, cosine, no cycle_momentum
119
+ - Regularization weight: 0.1
120
+
121
+ ## Prompt format
122
+
123
+ The format for reddit-instruct and oasst2 was:
124
+
125
+ ```
126
+ <|user|>
127
+ [insert instruction here]
128
+ <|assistant|>
129
+ [insert response here]
130
+ <|user|>
131
+ ...
132
+ ```
133
+
134
+ The format for TinyCoT was:
135
+ ```
136
+ <|user|>
137
+ [insert instruction here]
138
+ <|rationale|>
139
+ [insert reasoning here]
140
+ <|answer|>
141
+ [insert direct answer here]
142
+ ```