psxjp5 commited on
Commit
f8aa6e8
1 Parent(s): 03890a3

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +12 -6
README.md CHANGED
@@ -11,12 +11,9 @@ model-index:
11
  results: []
12
  ---
13
 
14
- <!-- This model card has been generated automatically according to the information the Trainer had access to. You
15
- should probably proofread and complete it, then remove this comment. -->
16
-
17
  # mt5-small_test_45
18
 
19
- This model is a fine-tuned version of [google/mt5-small](https://huggingface.co/google/mt5-small) on the None dataset.
20
  It achieves the following results on the evaluation set:
21
  - Loss: 0.7291
22
  - Rouge1: 44.4366
@@ -32,11 +29,20 @@ It achieves the following results on the evaluation set:
32
 
33
  ## Model description
34
 
35
- More information needed
 
 
 
 
36
 
37
  ## Intended uses & limitations
38
 
39
- More information needed
 
 
 
 
 
40
 
41
  ## Training and evaluation data
42
 
 
11
  results: []
12
  ---
13
 
 
 
 
14
  # mt5-small_test_45
15
 
16
+ This model is a fine-tuned version of [google/mt5-small](https://huggingface.co/google/mt5-small) on an enhanced version of the Natural Questions dataset.
17
  It achieves the following results on the evaluation set:
18
  - Loss: 0.7291
19
  - Rouge1: 44.4366
 
29
 
30
  ## Model description
31
 
32
+ This model is fine-tuned for long-form, closed-domain question answering - question-answering from context. It uses a heavily refined version of Google's Natural Questions dataset.
33
+
34
+ Answers to the questions were rewritten using OpenAI's GPT-3.5 Turbo model.
35
+
36
+ Please see [the following repo](https://github.com/pointonjoel/MSc-Diss) for all code and adaptations.
37
 
38
  ## Intended uses & limitations
39
 
40
+ The model requires questions to be submitted using the following format using the input message:
41
+ \[CONTEXT\] <\s> \[QUESTION\]
42
+
43
+ It is trained to respond appropriately when a question cannot be answered using the provided context.
44
+
45
+ It can give false negatives and false positives on occasion (see Training Results), and all answers must be checked appropriately.
46
 
47
  ## Training and evaluation data
48