Upload folder using huggingface_hub

.ipynb_checkpoints/README-checkpoint.md

@@ -0,0 +1,100 @@
+---
+language:
+- en
+- ru
+tags:
+- translation
+license: cc-by-4.0
+datasets:
+- quickmt/quickmt-train.ru-en
+model-index:
+- name: quickmt-ru-en
+  results:
+  - task:
+      name: Translation rus-eng
+      type: translation
+      args: rus-eng
+    dataset:
+      name: flores101-devtest
+      type: flores_101
+      args: rus_Cyrl eng_Latn devtest
+    metrics:
+    - name: BLEU
+      type: bleu
+      value: 33.9 
+    - name: CHRF
+      type: chrf
+      value:  61.63
+    - name: COMET
+      type: comet
+      value: 85.7
+---
+
+
+# `quickmt-ru-en` Neural Machine Translation Model 
+
+`quickmt-ru-en` is a reasonably fast and reasonably accurate neural machine translation model for translation from `ru` into `en`.
+
+
+## Model Information
+
+* Trained using [`eole`](https://github.com/eole-nlp/eole)
+* 185M parameter transformer 'big' with 8 encoder layers and 2 decoder layers
+* 50k joint Sentencepiece vocabulary
+* Exported for fast inference to [CTranslate2](https://github.com/OpenNMT/CTranslate2) format
+* Training data: https://huggingface.co/datasets/quickmt/quickmt-train.ru-en/tree/main
+
+See the `eole` model configuration in this repository for further details and the `eole-model` for the raw `eole` (pytorch) model.
+
+## Usage with `quickmt`
+
+You must install the Nvidia cuda toolkit first, if you want to do GPU inference.
+
+Next, install the `quickmt` python library and download the model:
+
+```bash
+git clone https://github.com/quickmt/quickmt.git
+pip install ./quickmt/
+
+quickmt-model-download quickmt/quickmt-ru-en ./quickmt-ru-en
+```
+
+Finally use the model in python:
+
+```python
+from quickmt import Translator
+
+# Auto-detects GPU, set to "cpu" to force CPU inference
+t = Translator("./quickmt-ru-en/", device="auto")
+
+# Translate - set beam size to 1 for faster speed (but lower quality)
+sample_text = 'Согласно предупреждению доктора Эхуда Ура (Ehud Ur), профессора медицины в Университете Дэлхаузи в Галифаксе (Новая Шотландия) и председателя клинико-научного отдела Канадской диабетической ассоциации, исследования все еще находятся на начальной стадии.'
+
+t(sample_text, beam_size=5)
+```
+
+> 'According to the warning of Dr. Ehud Ur, Professor of Medicine at Dalhousie University in Halifax, Nova Scotia, and Chair of the Clinical Science Division of the Canadian Diabetes Association, the research is still in its infancy.'
+
+```python
+# Get alternative translations by sampling
+# You can pass any cTranslate2 `translate_batch` arguments
+t([sample_text], sampling_temperature=1.2, beam_size=1, sampling_topk=50, sampling_topp=0.9)
+```
+
+> 'According to the warning of Professor Ehud Ur, a Professor of Medicine at Dalhousie University in Halifax, Nova Scotia, and Chair of the Clinical and Scientific Division of the Canadian Diabetes Association, research is still in a very early stage.'
+
+
+The model is in `ctranslate2` format, and the tokenizers are `sentencepiece`, so you can use `ctranslate2` directly instead of through `quickmt`. It is also possible  to get this model to work with e.g. [LibreTranslate](https://libretranslate.com/) which also uses `ctranslate2` and `sentencepiece`.
+
+## Metrics
+
+`bleu` and `chrf2` are calculated with [sacrebleu](https://github.com/mjpost/sacrebleu) on the [Flores200 `devtest` test set](https://huggingface.co/datasets/facebook/flores) ("rus_Cyrl"->"eng_Latn"). `comet22` with the [`comet`](https://github.com/Unbabel/COMET) library and the [default model](https://huggingface.co/Unbabel/wmt22-comet-da). "Time (s)" is the time in seconds to translate the flores-devtest dataset (1012 sentences) on an RTX 4070s GPU with batch size 32 (faster speed is possible using a larger batch size).
+
+|                                  |   bleu |   chrf2 |   comet22 |   Time (s) |
+|:---------------------------------|-------:|--------:|----------:|-----------:|
+| quickmt/quickmt-ru-en            |  33.9  |   61.63 |     85.7  |       1.31 |
+| Helsink-NLP/opus-mt-ru-en        |  30.04 |   58.23 |     83.97 |       3.72 |
+| facebook/nllb-200-distilled-600M |  34.59 |   61.26 |     85.88 |      21.93 |
+| facebook/nllb-200-distilled-1.3B |  36.99 |   63.04 |     86.59 |      38.12 |
+| facebook/m2m100_418M             |  26.62 |   56.31 |     81.77 |      18.73 |
+| facebook/m2m100_1.2B             |  32.01 |   60.3  |     85.01 |      35.99 |

README.md

@@ -1,3 +1,100 @@
----
-license: cc-by-4.0
----
+---
+language:
+- en
+- ru
+tags:
+- translation
+license: cc-by-4.0
+datasets:
+- quickmt/quickmt-train.ru-en
+model-index:
+- name: quickmt-ru-en
+  results:
+  - task:
+      name: Translation rus-eng
+      type: translation
+      args: rus-eng
+    dataset:
+      name: flores101-devtest
+      type: flores_101
+      args: rus_Cyrl eng_Latn devtest
+    metrics:
+    - name: BLEU
+      type: bleu
+      value: 33.9 
+    - name: CHRF
+      type: chrf
+      value:  61.63
+    - name: COMET
+      type: comet
+      value: 85.7
+---
+
+
+# `quickmt-ru-en` Neural Machine Translation Model 
+
+`quickmt-ru-en` is a reasonably fast and reasonably accurate neural machine translation model for translation from `ru` into `en`.
+
+
+## Model Information
+
+* Trained using [`eole`](https://github.com/eole-nlp/eole)
+* 185M parameter transformer 'big' with 8 encoder layers and 2 decoder layers
+* 50k joint Sentencepiece vocabulary
+* Exported for fast inference to [CTranslate2](https://github.com/OpenNMT/CTranslate2) format
+* Training data: https://huggingface.co/datasets/quickmt/quickmt-train.ru-en/tree/main
+
+See the `eole` model configuration in this repository for further details and the `eole-model` for the raw `eole` (pytorch) model.
+
+## Usage with `quickmt`
+
+You must install the Nvidia cuda toolkit first, if you want to do GPU inference.
+
+Next, install the `quickmt` python library and download the model:
+
+```bash
+git clone https://github.com/quickmt/quickmt.git
+pip install ./quickmt/
+
+quickmt-model-download quickmt/quickmt-ru-en ./quickmt-ru-en
+```
+
+Finally use the model in python:
+
+```python
+from quickmt import Translator
+
+# Auto-detects GPU, set to "cpu" to force CPU inference
+t = Translator("./quickmt-ru-en/", device="auto")
+
+# Translate - set beam size to 1 for faster speed (but lower quality)
+sample_text = 'Согласно предупреждению доктора Эхуда Ура (Ehud Ur), профессора медицины в Университете Дэлхаузи в Галифаксе (Новая Шотландия) и председателя клинико-научного отдела Канадской диабетической ассоциации, исследования все еще находятся на начальной стадии.'
+
+t(sample_text, beam_size=5)
+```
+
+> 'According to the warning of Dr. Ehud Ur, Professor of Medicine at Dalhousie University in Halifax, Nova Scotia, and Chair of the Clinical Science Division of the Canadian Diabetes Association, the research is still in its infancy.'
+
+```python
+# Get alternative translations by sampling
+# You can pass any cTranslate2 `translate_batch` arguments
+t([sample_text], sampling_temperature=1.2, beam_size=1, sampling_topk=50, sampling_topp=0.9)
+```
+
+> 'According to the warning of Professor Ehud Ur, a Professor of Medicine at Dalhousie University in Halifax, Nova Scotia, and Chair of the Clinical and Scientific Division of the Canadian Diabetes Association, research is still in a very early stage.'
+
+
+The model is in `ctranslate2` format, and the tokenizers are `sentencepiece`, so you can use `ctranslate2` directly instead of through `quickmt`. It is also possible  to get this model to work with e.g. [LibreTranslate](https://libretranslate.com/) which also uses `ctranslate2` and `sentencepiece`.
+
+## Metrics
+
+`bleu` and `chrf2` are calculated with [sacrebleu](https://github.com/mjpost/sacrebleu) on the [Flores200 `devtest` test set](https://huggingface.co/datasets/facebook/flores) ("rus_Cyrl"->"eng_Latn"). `comet22` with the [`comet`](https://github.com/Unbabel/COMET) library and the [default model](https://huggingface.co/Unbabel/wmt22-comet-da). "Time (s)" is the time in seconds to translate the flores-devtest dataset (1012 sentences) on an RTX 4070s GPU with batch size 32 (faster speed is possible using a larger batch size).
+
+|                                  |   bleu |   chrf2 |   comet22 |   Time (s) |
+|:---------------------------------|-------:|--------:|----------:|-----------:|
+| quickmt/quickmt-ru-en            |  33.9  |   61.63 |     85.7  |       1.31 |
+| Helsink-NLP/opus-mt-ru-en        |  30.04 |   58.23 |     83.97 |       3.72 |
+| facebook/nllb-200-distilled-600M |  34.59 |   61.26 |     85.88 |      21.93 |
+| facebook/nllb-200-distilled-1.3B |  36.99 |   63.04 |     86.59 |      38.12 |
+| facebook/m2m100_418M             |  26.62 |   56.31 |     81.77 |      18.73 |
+| facebook/m2m100_1.2B             |  32.01 |   60.3  |     85.01 |      35.99 |

config.json

@@ -0,0 +1,10 @@
+{
+  "add_source_bos": false,
+  "add_source_eos": false,
+  "bos_token": "<s>",
+  "decoder_start_token": "<s>",
+  "eos_token": "</s>",
+  "layer_norm_epsilon": 1e-06,
+  "multi_query_attention": false,
+  "unk_token": "<unk>"
+}

eole-config.yaml

@@ -0,0 +1,98 @@
+## IO
+save_data: data
+overwrite: True
+seed: 1234
+report_every: 100
+valid_metrics: ["BLEU"]
+tensorboard: true
+tensorboard_log_dir: tensorboard
+
+### Vocab
+src_vocab: ru.eole.vocab
+tgt_vocab: en.eole.vocab
+src_vocab_size: 20000
+tgt_vocab_size: 20000
+vocab_size_multiple: 8
+share_vocab: false
+n_sample: 0
+
+data:
+    corpus_1:
+        # path_src: hf://quickmt/quickmt-train.ru-en/ru
+        # path_tgt: hf://quickmt/quickmt-train.ru-en/en
+        # path_sco: hf://quickmt/quickmt-train.ru-en/sco
+        path_src: train.ru
+        path_tgt: train.en
+    valid:
+        path_src: dev.ru
+        path_tgt: dev.en
+
+transforms: [sentencepiece, filtertoolong]
+transforms_configs:
+  sentencepiece:
+    src_subword_model: "ru.spm.model"
+    tgt_subword_model: "en.spm.model"
+  filtertoolong:
+    src_seq_length: 256
+    tgt_seq_length: 256
+
+training:
+    # Run configuration
+    model_path: quickmt-ru-en-eole-model
+    #train_from: model
+    keep_checkpoint: 4
+    train_steps: 100000
+    save_checkpoint_steps: 5000
+    valid_steps: 5000
+    
+    # Train on a single GPU
+    world_size: 1
+    gpu_ranks: [0]
+
+    # Batching 10240
+    batch_type: "tokens"
+    batch_size: 8000
+    valid_batch_size: 4096
+    batch_size_multiple: 8
+    accum_count: [10]
+    accum_steps: [0]
+
+    # Optimizer & Compute
+    compute_dtype: "fp16"
+    optim: "adamw"
+    #use_amp: False
+    learning_rate: 2.0
+    warmup_steps: 4000
+    decay_method: "noam"
+    adam_beta2: 0.998
+
+    # Data loading
+    bucket_size: 128000
+    num_workers: 4
+    prefetch_factor: 32
+    
+    # Hyperparams
+    dropout_steps: [0]
+    dropout: [0.1]
+    attention_dropout: [0.1]
+    max_grad_norm: 0
+    label_smoothing: 0.1
+    average_decay: 0.0001
+    param_init_method: xavier_uniform
+    normalization: "tokens"
+
+model:
+    architecture: "transformer"
+    share_embeddings: false
+    share_decoder_embeddings: false
+    hidden_size: 1024
+    encoder:
+        layers: 8
+    decoder:
+        layers: 2
+    heads: 8
+    transformer_ff: 4096
+    embeddings:
+        word_vec_size: 1024
+        position_encoding_type: "SinusoidalInterleaved"
+

eole-model/config.json

@@ -0,0 +1,132 @@
+{
+  "report_every": 100,
+  "tgt_vocab": "en.eole.vocab",
+  "valid_metrics": [
+    "BLEU"
+  ],
+  "tensorboard": true,
+  "src_vocab": "ru.eole.vocab",
+  "transforms": [
+    "sentencepiece",
+    "filtertoolong"
+  ],
+  "vocab_size_multiple": 8,
+  "tensorboard_log_dir": "tensorboard",
+  "seed": 1234,
+  "n_sample": 0,
+  "save_data": "data",
+  "share_vocab": false,
+  "src_vocab_size": 20000,
+  "tensorboard_log_dir_dated": "tensorboard/May-06_17-28-49",
+  "tgt_vocab_size": 20000,
+  "overwrite": true,
+  "training": {
+    "bucket_size": 128000,
+    "dropout_steps": [
+      0
+    ],
+    "keep_checkpoint": 4,
+    "average_decay": 0.0001,
+    "param_init_method": "xavier_uniform",
+    "attention_dropout": [
+      0.1
+    ],
+    "train_steps": 100000,
+    "batch_size": 8000,
+    "accum_steps": [
+      0
+    ],
+    "prefetch_factor": 32,
+    "max_grad_norm": 0.0,
+    "valid_batch_size": 4096,
+    "dropout": [
+      0.1
+    ],
+    "num_workers": 0,
+    "decay_method": "noam",
+    "valid_steps": 5000,
+    "model_path": "quickmt-ru-en-eole-model",
+    "world_size": 1,
+    "learning_rate": 2.0,
+    "save_checkpoint_steps": 5000,
+    "optim": "adamw",
+    "normalization": "tokens",
+    "adam_beta2": 0.998,
+    "warmup_steps": 4000,
+    "batch_size_multiple": 8,
+    "label_smoothing": 0.1,
+    "compute_dtype": "torch.float16",
+    "gpu_ranks": [
+      0
+    ],
+    "accum_count": [
+      10
+    ],
+    "batch_type": "tokens"
+  },
+  "model": {
+    "transformer_ff": 4096,
+    "architecture": "transformer",
+    "hidden_size": 1024,
+    "share_decoder_embeddings": false,
+    "position_encoding_type": "SinusoidalInterleaved",
+    "heads": 8,
+    "share_embeddings": false,
+    "embeddings": {
+      "src_word_vec_size": 1024,
+      "position_encoding_type": "SinusoidalInterleaved",
+      "tgt_word_vec_size": 1024,
+      "word_vec_size": 1024
+    },
+    "encoder": {
+      "transformer_ff": 4096,
+      "hidden_size": 1024,
+      "layers": 8,
+      "position_encoding_type": "SinusoidalInterleaved",
+      "encoder_type": "transformer",
+      "src_word_vec_size": 1024,
+      "heads": 8,
+      "n_positions": null
+    },
+    "decoder": {
+      "transformer_ff": 4096,
+      "tgt_word_vec_size": 1024,
+      "decoder_type": "transformer",
+      "hidden_size": 1024,
+      "layers": 2,
+      "position_encoding_type": "SinusoidalInterleaved",
+      "heads": 8,
+      "n_positions": null
+    }
+  },
+  "transforms_configs": {
+    "filtertoolong": {
+      "tgt_seq_length": 256,
+      "src_seq_length": 256
+    },
+    "sentencepiece": {
+      "tgt_subword_model": "${MODEL_PATH}/en.spm.model",
+      "src_subword_model": "${MODEL_PATH}/ru.spm.model"
+    }
+  },
+  "data": {
+    "corpus_1": {
+      "path_src": "train.ru",
+      "transforms": [
+        "sentencepiece",
+        "filtertoolong"
+      ],
+      "path_align": null,
+      "path_tgt": "train.en"
+    },
+    "valid": {
+      "path_src": "dev.ru",
+      "transforms": [
+        "sentencepiece",
+        "filtertoolong"
+      ],
+      "path_align": null,
+      "path_tgt": "dev.en"
+    }
+  }
+}

eole-model/en.spm.model

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a02eda49b9085bb516903b522f5123f22a730b2be29337202ff8280b786a680e
+size 592273

eole-model/model.00.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8b26000560eac8419c39aa672d130421a03ef65a60bd66818bfeeea152f0799d
+size 823882912

eole-model/ru.spm.model

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b428315194a0eebf6e69c4424ea2e78a1c03e20982eccd4e6eb446ca128124ec
+size 730304

model.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0fb552c7074a60314b7d411a3f533320d79f1236c05534a2dc8e926888af24bd
+size 401699775

src.spm.model

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b428315194a0eebf6e69c4424ea2e78a1c03e20982eccd4e6eb446ca128124ec
+size 730304

tgt.spm.model

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a02eda49b9085bb516903b522f5123f22a730b2be29337202ff8280b786a680e
+size 592273