readme update #6

README.md

@@ -16,18 +16,11 @@ language:
 
 # kazRush-kk-ru
 
-kazRush-kk-ru is a translation model for translating from Kazakh to Russian.
-
-- **Model type:** t5
-- **License:** apache-2.0
+kazRush-kk-ru is a translation model for translating from Kazakh to Russian. The model was trained with randomly initialized weights based on the T5 configuration on the available open-source parallel data.
 
 ## Usage
 
-Using the model requires some packages to be installed.
-
-```bash
-pip install numpy==1.26.4 torch~=2.2.2 transformers~=4.39.2 sentencepiece~=0.2.0
-```
+Using the model requires `sentencepiece` library to be installed.
 
 After installing necessary dependencies the model can be run with the following code:  
 
@@ -35,13 +28,14 @@ After installing necessary dependencies the model can be run with the following
 from transformers import AutoModelForSeq2SeqLM, AutoTokenizer
 import torch
 
-model = AutoModelForSeq2SeqLM.from_pretrained('deepvk/kazRush-kk-ru')
-tokenizer = AutoTokenizer.from_pretrained('deepvk/kazRush-kk-ru')
+device = 'cuda'
+model = AutoModelForSeq2SeqLM.from_pretrained('deepvk/KazRush-ru-kk').to(device)
+tokenizer = AutoTokenizer.from_pretrained('deepvk/KazRush-ru-kk')
 
+@torch.inference_mode
 def generate(text, **kwargs):
-    inputs = tokenizer(text, return_tensors='pt').to('cuda')
-    with torch.no_grad():
-        hypotheses = model.generate(**inputs, num_beams=5, **kwargs)
+    inputs = tokenizer(text, return_tensors='pt').to(device)
+    hypotheses = model.generate(**inputs, num_beams=5, **kwargs)
     return tokenizer.decode(hypotheses[0], skip_special_tokens=True)
 
 print(generate("Анам жақтауды жуды."))
@@ -56,41 +50,39 @@ You can also access the model via _pipeline_ wrapper:
 [{'translation_text': 'Кто выпустил собак?'}]
 ```
 
-## Training Details
-
-### Training Data
+## Data and Training
 
 This model was trained on the following data (Russian-Kazakh language pairs):  
-[OPUS Corpora](<https://opus.nlpl.eu/results/ru&kk/corpus-result-table>)  
-[kazparc](<https://huggingface.co/datasets/issai/kazparc>)  
-[wmt19 dataset](<https://statmt.org/wmt19/translation-task.html#download>)
 
-#### Preprocessing
+| Dataset  | Number of pairs | 
+|-----------------------------------------|-------|
+| [OPUS Corpora](<https://opus.nlpl.eu/results/ru&kk/corpus-result-table>)     | 718K   |
+| [kazparc](<https://huggingface.co/datasets/issai/kazparc>)            | 2,150K   |
+| [wmt19 dataset](<https://statmt.org/wmt19/translation-task.html#download>)                  | 5,063K   |
+| [TIL dataset](<https://github.com/turkic-interlingua/til-mt/tree/master/til_corpus>)                  | 4,403K   |
 
 Preprocessing of the data included:
-- deduplication;
-- removing trash symbols, special tags, multiple whitespaces etc. from texts;
-- removing texts that were not in Russian or Kazakh (language detection was made via [fasttext](<https://huggingface.co/facebook/fasttext-language-identification>));
-- removing pairs that had low alingment score (comparison was performed via [LaBSE](<https://huggingface.co/sentence-transformers/LaBSE>));
-- filtering the data using [opusfilter](<https://github.com/Helsinki-NLP/OpusFilter>) tools.
-
-#### Training
+1. deduplication
+2. removing trash symbols, special tags, multiple whitespaces etc. from texts
+3. removing texts that were not in Russian or Kazakh (language detection was made via [facebook/fasttext-language-identification](<https://huggingface.co/facebook/fasttext-language-identification>))
+4. removing pairs that had low alingment score (comparison was performed via [sentence-transformers/LaBSE](<https://huggingface.co/sentence-transformers/LaBSE>))
+5. filtering the data using [opusfilter](<https://github.com/Helsinki-NLP/OpusFilter>) tools
 
 The model was trained for 56 hours on 2 GPUs NVIDIA A100 80 Gb.
 
 ## Evaluation
 
-Current model was compared to another open-source translation model, NLLB. We compared our model to all version of nllb, excluding nllb-moe-54b due to its size.
+Current model was compared to another open-source translation model, [NLLB](<https://huggingface.co/docs/transformers/model_doc/nllb>). We compared our model to all version of NLLB, excluding nllb-moe-54b due to its size.
 The metrics - BLEU, chrF and COMET - were calculated on `devtest` part of [FLORES+ evaluation benchmark](<https://github.com/openlanguagedata/flores>), most recent evaluation benchmark for multilingual machine translation.  
 Calculation of BLEU and chrF follows the standart implementation from [sacreBLEU](<https://github.com/mjpost/sacrebleu>), and COMET is calculated using default model described in [COMET repository](<https://github.com/Unbabel/COMET>).
 
 | Model  | Size | BLEU | chrf | comet |
 |-----------------------------------------|-------|-----------------------------|------------------------|----------|
-| [nllb-200-distilled-600M](https://huggingface.co/facebook/nllb-200-distilled-600M)    | 600M   | 18.0  |  47.3  | 0.8563 |
-| [our model (91ejx732)]()                             | 197 M    | 18.8      | 48.7                   | 0.8672 |
-| [nllb-200-1.3B](https://huggingface.co/facebook/nllb-200-1.3B)                   | 1.3B   | 20.4 | 49.3          | 0.8795 |
-| [nllb-200-distilled-1.3B](https://huggingface.co/facebook/nllb-200-distilled-1.3B)   | 1.3B    | 20.8 | 49.6 | 0.8814 |
-| [nllb-200-3.3B](https://huggingface.co/facebook/nllb-200-3.3B)                    | 3.3B    | 21.5  | 50.7  | 0.8874 |
+| [nllb-200-distilled-600M](https://huggingface.co/facebook/nllb-200-distilled-600M)    | 600M   | 18.0  |  47.3  | 85.6 |
+| [This model (91ejx732)]()                             | 197M    | 18.8      | 48.7                   | 86.7 |
+| [nllb-200-1.3B](https://huggingface.co/facebook/nllb-200-1.3B)                   | 1.3B   | 20.4 | 49.3          | 87.9 |
+| [nllb-200-distilled-1.3B](https://huggingface.co/facebook/nllb-200-distilled-1.3B)   | 1.3B    | 20.8 | 49.6 | 88.1 |
+| [nllb-200-3.3B](https://huggingface.co/facebook/nllb-200-3.3B)                    | 3.3B    | 21.5  | 50.7  | 88.7 |
 
 ## Examples of usage: