ptrdvn commited on
Commit
ac4252b
Β·
verified Β·
1 Parent(s): 060faf8

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +59 -1
README.md CHANGED
@@ -9,6 +9,49 @@ model-index:
9
  results: []
10
  ---
11
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
12
  <!-- This model card has been generated automatically according to the information the Trainer had access to. You
13
  should probably proofread and complete it, then remove this comment. -->
14
 
@@ -150,4 +193,19 @@ The following hyperparameters were used during training:
150
  - Transformers 4.38.2
151
  - Pytorch 2.2.1+cu121
152
  - Datasets 2.18.0
153
- - Tokenizers 0.15.0
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
9
  results: []
10
  ---
11
 
12
+ # Suzume ORPO
13
+
14
+ <p align="center">
15
+ <img width=500 src="https://cdn-uploads.huggingface.co/production/uploads/64b63f8ad57e02621dc93c8b/kWQSu02YfgYdUQqv4s5lq.png" alt="Suzume with Mitsu - a Japanese tree sparrow with honey on it"/>
16
+ </p>
17
+
18
+ [[Paper]](https://arxiv.org/abs/2405.18952) [[Dataset]](https://huggingface.co/datasets/lightblue/mitsu)
19
+
20
+ This is Suzume ORPO, an ORPO trained fine-tune of the [lightblue/suzume-llama-3-8B-multilingual](https://huggingface.co/lightblue/suzume-llama-3-8B-multilingual) model using our [lightblue/mitsu](https://huggingface.co/datasets/lightblue/mitsu) dataset.
21
+
22
+ We have trained several versions of this model using ORPO and so recommend that you use the best performing model from our tests, [lightblue/suzume-llama-3-8B-multilingual-orpo-borda-half](https://huggingface.co/lightblue/suzume-llama-3-8B-multilingual-orpo-borda-half).
23
+
24
+ Note that this model has a non-commerical license as we used the Command R and Command R+ models to generate our training data for this model ([lightblue/mitsu](https://huggingface.co/datasets/lightblue/mitsu)).
25
+
26
+ We are currently working on a developing a commerically usable model, so stay tuned for that!
27
+
28
+ # Model results
29
+
30
+ We compare the MT-Bench scores across 6 languages for our 4 ORPO trained models, as well as some baselines:
31
+
32
+ * [meta-llama/Meta-Llama-3-8B-Instruct](https://huggingface.co/meta-llama/Meta-Llama-3-8B-Instruct) - The foundation model that our models are ultimately built upon
33
+ * [Nexusflow/Starling-LM-7B-beta](https://huggingface.co/Nexusflow/Starling-LM-7B-beta) - The highest performing open model on the Chatbot arena that is of a similar size to ours
34
+ * gpt-3.5-turbo - A fairly high quality (although not state-of-the-art) proprietary LLM
35
+ * [lightblue/suzume-llama-3-8B-multilingual](https://huggingface.co/lightblue/suzume-llama-3-8B-multilingual) - The base model which we train our ORPO finetunes from
36
+
37
+ | **MT-Bench language** | **meta-llama/Meta-Llama-3-8B-Instruct** | **Nexusflow/Starling-LM-7B-beta** | **gpt-3.5-turbo** | **lightblue/suzume-llama-3-8B-multilingual** | **lightblue/suzume-llama-3-8B-multilingual-orpo-borda-full** | **lightblue/suzume-llama-3-8B-multilingual-orpo-borda-top75** | **lightblue/suzume-llama-3-8B-multilingual-orpo-borda-half** | **lightblue/suzume-llama-3-8B-multilingual-orpo-borda-top25** |
38
+ |-----------------------|-----------------------------------------|-----------------------------------|-------------------|----------------------------------------------|--------------------------------------------------------------|---------------------------------------------------------------|--------------------------------------------------------------|---------------------------------------------------------------|
39
+ | **Chinese πŸ‡¨πŸ‡³** | NaN | 6.97 | 7.55 | 7.11 | 7.65 | **7.77** | 7.74 | 7.44 |
40
+ | **English πŸ‡ΊπŸ‡Έ** | 7.98 | 7.92 | **8.26** | 7.73 | 7.98 | 7.94 | 7.98 | 8.22 |
41
+ | **French πŸ‡«πŸ‡·** | NaN | 7.29 | 7.74 | 7.66 | **7.84** | 7.46 | 7.78 | 7.81 |
42
+ | **German πŸ‡©πŸ‡ͺ** | NaN | 6.99 | 7.68 | 7.26 | 7.28 | 7.64 | 7.7 | **7.71** |
43
+ | **Japanese πŸ‡―πŸ‡΅** | NaN | 6.22 | **7.84** | 6.56 | 7.2 | 7.12 | 7.34 | 7.04 |
44
+ | **Russian πŸ‡·πŸ‡Ί** | NaN | 8.28 | 7.94 | 8.19 | 8.3 | 8.74 | **8.94** | 8.81 |
45
+
46
+ We can see noticable improvement on most languages compared to the base model. We also find that our ORPO models achieve the highest score out of all the models we evaluated for a number of languages.
47
+
48
+ # Training data
49
+
50
+ We trained this model using the [lightblue/mitsu_full_borda](https://huggingface.co/datasets/lightblue/mitsu_full_borda) dataset.
51
+
52
+ # Training configuration
53
+
54
+
55
  <!-- This model card has been generated automatically according to the information the Trainer had access to. You
56
  should probably proofread and complete it, then remove this comment. -->
57
 
 
193
  - Transformers 4.38.2
194
  - Pytorch 2.2.1+cu121
195
  - Datasets 2.18.0
196
+ - Tokenizers 0.15.0
197
+
198
+ # How to cite
199
+
200
+ ```tex
201
+ @article{devine2024sure,
202
+ title={Are You Sure? Rank Them Again: Repeated Ranking For Better Preference Datasets},
203
+ author={Devine, Peter},
204
+ journal={arXiv preprint arXiv:2405.18952},
205
+ year={2024}
206
+ }
207
+ ```
208
+
209
+ # Developer
210
+
211
+ Peter Devine - ([ptrdvn](https://huggingface.co/ptrdvn))