zouharvi commited on
Commit
c5ec35e
·
verified ·
1 Parent(s): cecb748

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +27 -8
README.md CHANGED
@@ -101,7 +101,7 @@ base_model:
101
  - FacebookAI/xlm-roberta-large
102
  ---
103
 
104
- # COMET-poly-ic1-wmt25
105
 
106
  This model is based on [COMET-poly](https://github.com/zouharvi/COMET-poly), which is a fork but not compatible with original Unbabel's COMET.
107
  To run the model, you need to first install this version of COMET either with:
@@ -115,38 +115,57 @@ cd COMET-poly
115
  pip3 install -e comet_poly
116
  ```
117
 
118
- This model scores the translation `mt` but takes additional in-context example: source `src2`, translation `mt2`, and score `score2`, which makes it a better quality estimator:
119
  ```python
120
  import comet_poly
121
- model = comet_poly.load_from_checkpoint(comet_poly.download_model("zouharvi/COMET-poly-ic1-wmt25"))
122
  data = [
123
  {
124
  "src": "Iceberg lettuce got its name in the 1920s when it was shipped packed in ice to stay fresh.",
125
  "mt": "Eisbergsalat erhielt seinen Namen in den 1920er-Jahren, als er in Eis verpackt verschickt wurde, um frisch zu bleiben.",
126
  "src2": "Lettuce is mostly water, which helps keep it crisp when chilled.",
127
  "mt2": "Kopfsalat besteht größtenteils aus Wasser, was ihm hilft, beim Kühlen knackig zu bleiben.",
128
- "score2": 94.5
 
 
 
 
 
 
129
  },
130
  {
131
  "src": "Goats have rectangular pupils, which give them a wide field of vision—up to 320 degrees!",
132
  "mt": "Kozy mají obdélníkové zornice, což jim umožňuje vidět skoro všude kolem sebe, aniž by musely otáčet hlavou.",
133
  "src2": "Sheep, like goats, also have rectangular pupils for better peripheral vision.",
134
  "mt2": "Вівці, як і кози, також мають прямокутні зіниці для кращого периферичного зору.",
135
- "score2": 96.0
 
 
 
 
 
 
136
  },
137
  {
138
  "src": "This helps them spot predators from almost all directions without moving their heads.",
139
  "mt": "Điều này giúp chúng phát hiện kẻ săn mồi từ gần như mọi hướng mà không cần quay đầu.",
140
  "src2": "Many prey animals have evolved to detect threats with minimal movement.",
141
  "mt2": "Nhiều động vật thịt có tiến hóa để xem mối nguy bằng nhỏ đi lại.",
142
- "score2": 42.3
 
 
 
 
 
 
143
  }
144
  ]
 
145
  print("scores", model.predict(data, batch_size=8, gpus=1).scores)
146
  ```
147
  Outputs:
148
  ```
149
- scores [98.09857940673828, 85.52458953857422, 83.38972473144531]
150
  ```
151
 
152
  You can use a readily-available training data to do the on-the-fly retrieval.
@@ -173,7 +192,7 @@ data_kb = list(datasets.load_dataset("zouharvi/wmt-human-all", split="train"))
173
  data_retrieved = comet_poly.retrieval.retrieve_from_kb(
174
  data=data,
175
  data_kb=data_kb,
176
- k=1,
177
  prevent_hardmatch=False,
178
  key="src",
179
  )
 
101
  - FacebookAI/xlm-roberta-large
102
  ---
103
 
104
+ # COMET-poly-ic3-wmt25
105
 
106
  This model is based on [COMET-poly](https://github.com/zouharvi/COMET-poly), which is a fork but not compatible with original Unbabel's COMET.
107
  To run the model, you need to first install this version of COMET either with:
 
115
  pip3 install -e comet_poly
116
  ```
117
 
118
+ This model scores the translation `mt` but takes additional three in-context example: sources `src2`, `src3`, `src4`, translations `mt2`, `mt3`, `mt4`, and scores `score2`, `score3`, `score4`, which makes it a better quality estimator:
119
  ```python
120
  import comet_poly
121
+ model = comet_poly.load_from_checkpoint(comet_poly.download_model("zouharvi/COMET-poly-ic3-wmt25"))
122
  data = [
123
  {
124
  "src": "Iceberg lettuce got its name in the 1920s when it was shipped packed in ice to stay fresh.",
125
  "mt": "Eisbergsalat erhielt seinen Namen in den 1920er-Jahren, als er in Eis verpackt verschickt wurde, um frisch zu bleiben.",
126
  "src2": "Lettuce is mostly water, which helps keep it crisp when chilled.",
127
  "mt2": "Kopfsalat besteht größtenteils aus Wasser, was ihm hilft, beim Kühlen knackig zu bleiben.",
128
+ "score2": 94.5,
129
+ "src3": "Iceberg lettuce is often used in burgers for its crunch and mild flavor.",
130
+ "mt3": "Íssalat er oft notað í hamborgara vegna stökkleika og milds bragðs.",
131
+ "score3": 92.0,
132
+ "src4": "Farmers harvest lettuce early in the morning to keep it fresh longer.",
133
+ "mt4": "Les agriculteurs récoltent la laitue tôt le matin pour la garder fraîche plus longtemps.",
134
+ "score4": 82.3
135
  },
136
  {
137
  "src": "Goats have rectangular pupils, which give them a wide field of vision—up to 320 degrees!",
138
  "mt": "Kozy mají obdélníkové zornice, což jim umožňuje vidět skoro všude kolem sebe, aniž by musely otáčet hlavou.",
139
  "src2": "Sheep, like goats, also have rectangular pupils for better peripheral vision.",
140
  "mt2": "Вівці, як і кози, також мають прямокутні зіниці для кращого периферичного зору.",
141
+ "score2": 96.0,
142
+ "src3": "This unique eye shape helps them detect predators early.",
143
+ "mt3": "Ця унікальна форма очей допомагає їм рано виявляти хижаків.",
144
+ "score3": 93.2,
145
+ "src4": "Goats' vision stays stable even when they climb steep surfaces.",
146
+ "mt4": "Kozí vidění zůstává stabilní když jdou na vysoký svah.",
147
+ "score4": 67.4
148
  },
149
  {
150
  "src": "This helps them spot predators from almost all directions without moving their heads.",
151
  "mt": "Điều này giúp chúng phát hiện kẻ săn mồi từ gần như mọi hướng mà không cần quay đầu.",
152
  "src2": "Many prey animals have evolved to detect threats with minimal movement.",
153
  "mt2": "Nhiều động vật thịt có tiến hóa để xem mối nguy bằng nhỏ đi lại.",
154
+ "score2": 42.3,
155
+ "src3": "Nepohybující se držení těla pomáhá zvířatům zůstat bez povšimnutí a přitom být ostražitá.",
156
+ "mt3": "Нерухома поза допомагає тваринам залишатися непоміченими, залишаючись при цьому пильними.",
157
+ "score3": 90.1,
158
+ "src4": "Animals like deer rely on peripheral vision to detect danger.",
159
+ "mt4": "Los animales como los ciervos confían en la visión periférica para detectar el peligro.",
160
+ "score4": 85.0
161
  }
162
  ]
163
+
164
  print("scores", model.predict(data, batch_size=8, gpus=1).scores)
165
  ```
166
  Outputs:
167
  ```
168
+ scores [98.42459869384766, 94.70307922363281, 91.14827728271484]
169
  ```
170
 
171
  You can use a readily-available training data to do the on-the-fly retrieval.
 
192
  data_retrieved = comet_poly.retrieval.retrieve_from_kb(
193
  data=data,
194
  data_kb=data_kb,
195
+ k=3, # this model takes three in-context examples
196
  prevent_hardmatch=False,
197
  key="src",
198
  )