Upload folder using huggingface_hub

.ipynb_checkpoints/EADME-checkpoint.md

@@ -0,0 +1,116 @@
+---
+language:
+- ar
+- en
+tags:
+- translation
+license: cc-by-4.0
+datasets:
+- quickmt/quickmt-train.ar-en
+- quickmt/madlad400-en-backtranslated-ar
+- quickmt/newscrawl2024-en-backtranslated-ar
+model-index:
+- name: quickmt-ar-en
+  results:
+  - task:
+      name: Translation arb-eng
+      type: translation
+      args: arb-eng
+    dataset:
+      name: flores101-devtest
+      type: flores_101
+      args: arb_Arab eng_Latn devtest
+    metrics:
+    - name: BLEU
+      type: bleu
+      value: 44.11
+    - name: CHRF
+      type: chrf
+      value: 67.96
+    - name: COMET
+      type: comet
+      value: 87.64
+---
+
+
+# `quickmt-ar-en` Neural Machine Translation Model 
+
+`quickmt-ar-en` is a reasonably fast and reasonably accurate neural machine translation model for translation from `ar` into `en`.
+
+`quickmt` models are roughly 3 times faster for GPU inference than OpusMT models and roughly [40 times](https://huggingface.co/spaces/quickmt/quickmt-vs-libretranslate) faster than [LibreTranslate](https://huggingface.co/spaces/quickmt/quickmt-vs-libretranslate)/[ArgosTranslate](github.com/argosopentech/argos-translate). 
+
+
+## *UPDATED VERSION!* 
+
+This model was trained with back-translated data and has improved translation quality!
+
+
+## Try it on our Huggingface Space
+
+Give it a try before downloading here: https://huggingface.co/spaces/quickmt/QuickMT-Demo
+
+
+## Model Information
+
+* Trained using [`eole`](https://github.com/eole-nlp/eole) 
+* 200M parameter seq2seq transformer
+* 32k separate Sentencepiece vocabs
+* Exported for fast inference to [CTranslate2](https://github.com/OpenNMT/CTranslate2) format
+* The pytorch model (for use with [`eole`](https://github.com/eole-nlp/eole)) is available in this repository in the `eole-model` folder
+
+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 -e ./quickmt/
+
+quickmt-model-download quickmt/quickmt-ar-en ./quickmt-ar-en
+```
+
+Finally use the model in python:
+
+```python
+from quickmt import Translator
+
+# Auto-detects GPU, set to "cpu" to force CPU inference
+mt = Translator("./quickmt-ar-en/", device="auto")
+
+# Translate - set beam size to 1 for faster speed (but lower quality)
+sample_text = 'نبه الدكتور إيهود أور -أستاذ الطب في جامعة دالهوزي في هاليفاكس، نوفا سكوتيا ورئيس الشعبة الطبية والعلمية في الجمعية الكندية للسكري- إلى أن البحث لا يزال في أيامه الأولى.'
+
+mt(sample_text, beam_size=5)
+```
+
+> 'Dr. Ehud Orr, a professor of medicine at Dalhousie University in Halifax, Nova Scotia and head of the medical and scientific division of the Canadian Diabetes Association, warned that the research is still in its early days.'
+
+```python
+# Get alternative translations by sampling
+# You can pass any cTranslate2 `translate_batch` arguments
+mt([sample_text], sampling_temperature=1.2, beam_size=1, sampling_topk=50, sampling_topp=0.9)
+```
+
+> 'Dr. Ehr, professor of medicine at Dalhousie University in Halifax, Nova Scotia and chief of the medical and scientific division at the Canadian Diabetes Association, warned that the research is still in its early days.'
+
+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`. A model in safetensors format to be used with `eole` is also provided.
+
+
+## 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). `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.
+
+
+|                                  |   bleu |   chrf2 |   comet22 |   Time (s) |
+|:---------------------------------|-------:|--------:|----------:|-----------:|
+| quickmt/quickmt-ar-en            |  44.11 |   67.96 |     87.64 |       1.11 |
+| Helsinki-NLP/opus-mt-ar-en       |  34.22 |   61.26 |     84.5  |       3.67 |
+| facebook/nllb-200-distilled-600M |  39.13 |   64.14 |     86.22 |      21.76 |
+| facebook/nllb-200-distilled-1.3B |  42.29 |   66.55 |     87.55 |      37.7  |
+| facebook/m2m100_418M             |  29.41 |   57.68 |     82.21 |      18.53 |
+| facebook/m2m100_1.2B             |  29.77 |   56.7  |     80.77 |      36.23 |

README.md

@@ -1,33 +1,35 @@
 ---
 language:
-- en
 - ar
+- en
 tags:
 - translation
 license: cc-by-4.0
 datasets:
 - quickmt/quickmt-train.ar-en
+- quickmt/madlad400-en-backtranslated-ar
+- quickmt/newscrawl2024-en-backtranslated-ar
 model-index:
 - name: quickmt-ar-en
   results:
   - task:
-      name: Translation ara-eng
+      name: Translation arb-eng
       type: translation
-      args: ara-eng
+      args: arb-eng
     dataset:
       name: flores101-devtest
       type: flores_101
-      args: ara_Arab eng_Latn devtest
+      args: arb_Arab eng_Latn devtest
     metrics:
-    - name: CHRF
-      type: chrf
-      value: 66.98
     - name: BLEU
       type: bleu
-      value: 42.79
+      value: 44.11
+    - name: CHRF
+      type: chrf
+      value: 67.96
     - name: COMET
       type: comet
-      value: 87.4
+      value: 87.64
 ---
 
 
@@ -35,14 +37,26 @@ model-index:
 
 `quickmt-ar-en` is a reasonably fast and reasonably accurate neural machine translation model for translation from `ar` into `en`.
 
+`quickmt` models are roughly 3 times faster for GPU inference than OpusMT models and roughly [40 times](https://huggingface.co/spaces/quickmt/quickmt-vs-libretranslate) faster than [LibreTranslate](https://huggingface.co/spaces/quickmt/quickmt-vs-libretranslate)/[ArgosTranslate](github.com/argosopentech/argos-translate). 
+
+
+## *UPDATED VERSION!* 
+
+This model was trained with back-translated data and has improved translation quality!
+
+
+## Try it on our Huggingface Space
+
+Give it a try before downloading here: https://huggingface.co/spaces/quickmt/QuickMT-Demo
+
 
 ## Model Information
 
-* Trained using [`eole`](https://github.com/eole-nlp/eole)
-* 185M parameter transformer 'big' with 8 encoder layers and 2 decoder layers
-* 20k sentencepiece vocabularies
+* Trained using [`eole`](https://github.com/eole-nlp/eole) 
+* 200M parameter seq2seq transformer
+* 32k separate Sentencepiece vocabs
 * Exported for fast inference to [CTranslate2](https://github.com/OpenNMT/CTranslate2) format
-* Training data: https://huggingface.co/datasets/quickmt/quickmt-train.ar-en/tree/main
+* The pytorch model (for use with [`eole`](https://github.com/eole-nlp/eole)) is available in this repository in the `eole-model` folder
 
 See the `eole` model configuration in this repository for further details and the `eole-model` for the raw `eole` (pytorch) model.
 
@@ -55,7 +69,7 @@ Next, install the `quickmt` python library and download the model:
 
 ```bash
 git clone https://github.com/quickmt/quickmt.git
-pip install ./quickmt/
+pip install -e ./quickmt/
 
 quickmt-model-download quickmt/quickmt-ar-en ./quickmt-ar-en
 ```
@@ -66,35 +80,37 @@ Finally use the model in python:
 from quickmt import Translator
 
 # Auto-detects GPU, set to "cpu" to force CPU inference
-t = Translator("./quickmt-ar-en/", device="auto")
+mt = Translator("./quickmt-ar-en/", device="auto")
 
-# Translate - set beam size to 5 for higher quality (but slower speed)
+# Translate - set beam size to 1 for faster speed (but lower quality)
 sample_text = 'نبه الدكتور إيهود أور -أستاذ الطب في جامعة دالهوزي في هاليفاكس، نوفا سكوتيا ورئيس الشعبة الطبية والعلمية في الجمعية الكندية للسكري- إلى أن البحث لا يزال في أيامه الأولى.'
-t(sample_text, beam_size=5)
 
-> 'Dr. Ehud Orr, professor of medicine at Dalhousie University in Halifax, Nova Scotia and head of the medical and scientific division of the Canadian Diabetes Association, warned that the research is still in its early days.'
+mt(sample_text, beam_size=5)
+```
+
+> 'Dr. Ehud Orr, a professor of medicine at Dalhousie University in Halifax, Nova Scotia and head of the medical and scientific division of the Canadian Diabetes Association, warned that the research is still in its early days.'
 
+```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)
-
-> 'Professor of Medicine at Dalhousie University in Halifax, Nova Scotia and chairman of the Medical and Scientific Division at the Canadian Diabetes Society, cautioned that the research was still in its early days.'
+mt([sample_text], sampling_temperature=1.2, beam_size=1, sampling_topk=50, sampling_topp=0.9)
 ```
 
-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`.
+> 'Dr. Ehr, professor of medicine at Dalhousie University in Halifax, Nova Scotia and chief of the medical and scientific division at the Canadian Diabetes Association, warned that the research is still in its early days.'
+
+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`. A model in safetensors format to be used with `eole` is also provided.
 
 
 ## 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) ("ara_Arab"->"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 (using `ctranslate2`) the flores-devtest dataset (1012 sentences) on an RTX 4070s GPU with batch size 32 (faster speed is possible using a large batch size).
+`bleu` and `chrf2` are calculated with [sacrebleu](https://github.com/mjpost/sacrebleu) on the [Flores200 `devtest` test set](https://huggingface.co/datasets/facebook/flores). `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.
+
 
 |                                  |   bleu |   chrf2 |   comet22 |   Time (s) |
 |:---------------------------------|-------:|--------:|----------:|-----------:|
-| quickmt/quickmt-ar-en            |  42.79 |   66.98 |     87.4  |       0.88 |
-| Helsink-NLP/opus-mt-ar-en        |  34.22 |   61.26 |     84.5  |       3.78 |
-| facebook/nllb-200-distilled-600M |  39.13 |   64.14 |     86.22 |      21.58 |
+| quickmt/quickmt-ar-en            |  44.11 |   67.96 |     87.64 |       1.11 |
+| Helsinki-NLP/opus-mt-ar-en       |  34.22 |   61.26 |     84.5  |       3.67 |
+| facebook/nllb-200-distilled-600M |  39.13 |   64.14 |     86.22 |      21.76 |
 | facebook/nllb-200-distilled-1.3B |  42.29 |   66.55 |     87.55 |      37.7  |
-| facebook/m2m100_418M             |  29.41 |   57.68 |     82.21 |      18.5  |
+| facebook/m2m100_418M             |  29.41 |   57.68 |     82.21 |      18.53 |
 | facebook/m2m100_1.2B             |  29.77 |   56.7  |     80.77 |      36.23 |
-
-`quickmt-ar-en` is the fastest and highest quality.

eole-config.yaml

@@ -1,5 +1,5 @@
 ## IO
-save_data: ar-en/data
+save_data: data
 overwrite: True
 seed: 1234
 report_every: 100
@@ -10,72 +10,100 @@ tensorboard_log_dir: tensorboard
 ### Vocab
 src_vocab: ar.eole.vocab
 tgt_vocab: en.eole.vocab
-src_vocab_size: 20000
-tgt_vocab_size: 20000
+src_vocab_size: 32000
+tgt_vocab_size: 32000
 vocab_size_multiple: 8
 share_vocab: false
 n_sample: 0
 
 data:
     corpus_1:
-        path_src: hf://quickmt/quickmt-train.ar-en/ar
-        path_tgt: hf://quickmt/quickmt-train.ar-en/en
-        path_sco: hf://quickmt/quickmt-train.ar-en/sco
+        path_src: train.ar
+        path_tgt: train.en
+        weight: 2
+    corpus_2:
+        path_src: newscrawl.backtrans.ar
+        path_tgt: newscrawl.2024.en
+        weight: 1
+    corpus_3:
+        path_src: madlad.backtrans.ar
+        path_tgt: madlad.en
+        weight: 2
     valid:
-        path_src: flores-dev.ar
-        path_tgt: flores-dev.en
+        path_src: valid.ar
+        path_tgt: valid.en
+
+# data:
+#     corpus_1:
+#         path_src: hf://quickmt/quickmt-train.ar-en/ar
+#         path_tgt: hf://quickmt/quickmt-train.ar-en/en
+#         path_sco: hf://quickmt/quickmt-train.ar-en/sco
+#         weight: 2
+#     corpus_2:
+#         path_src: hf://quickmt/newscrawl2024-en-backtranslated-ar/ar
+#         path_tgt: hf://quickmt/newscrawl2024-en-backtranslated-ar/en
+#         path_sco: hf://quickmt/newscrawl2024-en-backtranslated-ar/sco
+#         weight: 1
+#     corpus_3:
+#         path_src: hf://quickmt/madlad400-en-backtranslated-ar/ar
+#         path_tgt: hf://quickmt/madlad400-en-backtranslated-ar/en
+#         path_sco: hf://quickmt/madlad400-en-backtranslated-ar/sco
+#         weight: 2
+#     valid:
+#         path_src: valid.ar
+#         path_tgt: valid.en
+
+
 
 transforms: [sentencepiece, filtertoolong]
 transforms_configs:
   sentencepiece:
     src_subword_model: "ar.spm.model"
     tgt_subword_model: "en.spm.model"
-    src_subword_alpha: 0.5
-    src_subword_nbest: -1
   filtertoolong:
     src_seq_length: 256
     tgt_seq_length: 256
 
 training:
     # Run configuration
-    model_path: model
-    train_from: model
+    model_path: quickmt-ar-en-eole-model
     keep_checkpoint: 4
-    save_checkpoint_steps: 1000
     train_steps: 200000
-    valid_steps: 1000
+    save_checkpoint_steps: 5000
+    valid_steps: 5000
     
     # Train on a single GPU
     world_size: 1
     gpu_ranks: [0]
 
-    # Batching
+    # Batching 120,000 tokens
+    # For RTX 5090, 15000 batch size, accum_count 8
     batch_type: "tokens"
-    batch_size: 8192
-    valid_batch_size: 8192
+    batch_size: 15000
+    valid_batch_size: 2048
     batch_size_multiple: 8
-    accum_count: [16]
+    accum_count: [8]
     accum_steps: [0]
 
     # Optimizer & Compute
     compute_dtype: "fp16"
-    #use_amp: true
-    optim: "pagedadamw8bit"
-    learning_rate: 2.0
+    optim: "adamw"
+    use_amp: True
+    learning_rate: 3.0
     warmup_steps: 5000
     decay_method: "noam"
     adam_beta2: 0.998
 
     # Data loading
-    bucket_size: 128000
+    bucket_size: 256000
     num_workers: 4
-    prefetch_factor: 32
+    prefetch_factor: 128
     
     # Hyperparams
     dropout_steps: [0]
     dropout: [0.1]
-    attention_dropout: [0]
-    max_grad_norm: 2
+    attention_dropout: [0.1]
+    max_grad_norm: 0
     label_smoothing: 0.1
     average_decay: 0.0001
     param_init_method: xavier_uniform
@@ -83,22 +111,21 @@ training:
 
 model:
     architecture: "transformer"
-    layer_norm: standard
     share_embeddings: false
-    share_decoder_embeddings: true
-    add_ffnbias: true
-    mlp_activation_fn: gelu
+    share_decoder_embeddings: false
     add_estimator: false
+    add_ffnbias: true
     add_qkvbias: false
-    norm_eps: 1e-6
-    hidden_size: 1024
+    layer_norm: standard
+    mlp_activation_fn: gelu
+    hidden_size: 768
     encoder:
-        layers: 8
+        layers: 12
     decoder:
         layers: 2
-    heads: 8
+    heads: 16
     transformer_ff: 4096
     embeddings:
-        word_vec_size: 1024
+        word_vec_size: 768
         position_encoding_type: "SinusoidalInterleaved"
 

eole-model/ar.spm.model

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:c7b8bf117088a901628696f60ddd15e047057ccbdcf9b0996ca19e2779a234c7
-size 642429
+oid sha256:1a510ccfeb5cf06584327b9c2482ed0a4111d06348a56e28d165165b6dabe520
+size 901383

eole-model/config.json

@@ -1,152 +1,167 @@
 {
+  "seed": 1234,
+  "save_data": "data",
+  "report_every": 100,
+  "src_vocab_size": 32000,
   "src_vocab": "ar.eole.vocab",
-  "tensorboard_log_dir_dated": "tensorboard/Mar-21_20-28-42",
-  "tensorboard": true,
+  "overwrite": true,
+  "transforms": [
+    "sentencepiece",
+    "filtertoolong"
+  ],
   "valid_metrics": [
     "BLEU"
   ],
-  "save_data": "ar-en/data",
+  "tensorboard": true,
+  "tensorboard_log_dir_dated": "tensorboard/Dec-29_03-28-03",
+  "tgt_vocab": "en.eole.vocab",
   "vocab_size_multiple": 8,
+  "n_sample": 0,
   "share_vocab": false,
+  "tgt_vocab_size": 32000,
   "tensorboard_log_dir": "tensorboard",
-  "tgt_vocab_size": 20000,
-  "src_vocab_size": 20000,
-  "report_every": 100,
-  "transforms": [
-    "sentencepiece",
-    "filtertoolong"
-  ],
-  "seed": 1234,
-  "n_sample": 0,
-  "tgt_vocab": "en.eole.vocab",
-  "overwrite": true,
   "training": {
-    "batch_size": 8192,
-    "batch_type": "tokens",
-    "valid_steps": 1000,
-    "decay_method": "noam",
-    "optim": "pagedadamw8bit",
-    "bucket_size": 128000,
-    "keep_checkpoint": 4,
-    "model_path": "model",
+    "accum_count": [
+      8
+    ],
     "label_smoothing": 0.1,
-    "batch_size_multiple": 8,
-    "param_init_method": "xavier_uniform",
     "gpu_ranks": [
       0
     ],
-    "warmup_steps": 5000,
-    "prefetch_factor": 32,
-    "dropout": [
-      0.1
-    ],
-    "accum_steps": [
-      0
-    ],
+    "normalization": "tokens",
+    "average_decay": 0.0001,
+    "train_steps": 200000,
+    "prefetch_factor": 128,
+    "use_amp": true,
+    "param_init_method": "xavier_uniform",
+    "batch_size_multiple": 8,
+    "learning_rate": 3.0,
+    "adam_beta2": 0.998,
+    "batch_size": 15000,
     "world_size": 1,
     "attention_dropout": [
-      0.0
+      0.1
     ],
-    "average_decay": 0.0001,
-    "max_grad_norm": 2.0,
-    "learning_rate": 2.0,
-    "normalization": "tokens",
-    "valid_batch_size": 8192,
     "dropout_steps": [
       0
     ],
-    "num_workers": 0,
-    "accum_count": [
-      16
-    ],
+    "valid_steps": 5000,
     "compute_dtype": "torch.float16",
-    "adam_beta2": 0.998,
-    "save_checkpoint_steps": 1000,
-    "train_steps": 200000,
-    "train_from": "model"
+    "model_path": "quickmt-ar-en-eole-model",
+    "batch_type": "tokens",
+    "decay_method": "noam",
+    "dropout": [
+      0.1
+    ],
+    "warmup_steps": 5000,
+    "max_grad_norm": 0.0,
+    "num_workers": 0,
+    "save_checkpoint_steps": 5000,
+    "bucket_size": 256000,
+    "keep_checkpoint": 4,
+    "optim": "adamw",
+    "valid_batch_size": 2048,
+    "accum_steps": [
+      0
+    ]
   },
-  "transforms_configs": {
-    "filtertoolong": {
-      "tgt_seq_length": 256,
-      "src_seq_length": 256
+  "data": {
+    "corpus_1": {
+      "transforms": [
+        "sentencepiece",
+        "filtertoolong"
+      ],
+      "path_align": null,
+      "path_src": "train.ar",
+      "path_tgt": "train.en",
+      "weight": 2
     },
-    "sentencepiece": {
-      "tgt_subword_model": "${MODEL_PATH}/en.spm.model",
-      "src_subword_alpha": 0.5,
-      "src_subword_nbest": -1,
-      "src_subword_model": "${MODEL_PATH}/ar.spm.model"
+    "corpus_2": {
+      "transforms": [
+        "sentencepiece",
+        "filtertoolong"
+      ],
+      "path_align": null,
+      "path_src": "newscrawl.backtrans.ar",
+      "path_tgt": "newscrawl.2024.en",
+      "weight": 1
+    },
+    "corpus_3": {
+      "transforms": [
+        "sentencepiece",
+        "filtertoolong"
+      ],
+      "path_align": null,
+      "path_src": "madlad.backtrans.ar",
+      "path_tgt": "madlad.en",
+      "weight": 2
+    },
+    "valid": {
+      "transforms": [
+        "sentencepiece",
+        "filtertoolong"
+      ],
+      "path_align": null,
+      "path_src": "valid.ar",
+      "path_tgt": "valid.en"
     }
   },
   "model": {
-    "heads": 8,
-    "add_qkvbias": false,
-    "layer_norm": "standard",
     "share_embeddings": false,
-    "share_decoder_embeddings": true,
-    "position_encoding_type": "SinusoidalInterleaved",
-    "architecture": "transformer",
-    "hidden_size": 1024,
-    "add_ffnbias": true,
+    "heads": 16,
     "add_estimator": false,
-    "transformer_ff": 4096,
-    "norm_eps": 1e-06,
     "mlp_activation_fn": "gelu",
+    "transformer_ff": 4096,
+    "add_ffnbias": true,
+    "architecture": "transformer",
+    "layer_norm": "standard",
+    "add_qkvbias": false,
+    "hidden_size": 768,
+    "share_decoder_embeddings": false,
+    "position_encoding_type": "SinusoidalInterleaved",
     "embeddings": {
-      "src_word_vec_size": 1024,
-      "tgt_word_vec_size": 1024,
-      "word_vec_size": 1024,
-      "position_encoding_type": "SinusoidalInterleaved"
+      "word_vec_size": 768,
+      "src_word_vec_size": 768,
+      "position_encoding_type": "SinusoidalInterleaved",
+      "tgt_word_vec_size": 768
     },
-    "encoder": {
-      "heads": 8,
+    "decoder": {
+      "heads": 16,
+      "tgt_word_vec_size": 768,
+      "mlp_activation_fn": "gelu",
       "transformer_ff": 4096,
-      "hidden_size": 1024,
+      "add_ffnbias": true,
       "add_qkvbias": false,
       "layer_norm": "standard",
+      "hidden_size": 768,
+      "layers": 2,
+      "position_encoding_type": "SinusoidalInterleaved",
       "n_positions": null,
-      "add_ffnbias": true,
+      "decoder_type": "transformer"
+    },
+    "encoder": {
+      "heads": 16,
       "encoder_type": "transformer",
       "mlp_activation_fn": "gelu",
-      "layers": 8,
-      "src_word_vec_size": 1024,
-      "norm_eps": 1e-06,
-      "position_encoding_type": "SinusoidalInterleaved"
-    },
-    "decoder": {
-      "heads": 8,
-      "tgt_word_vec_size": 1024,
       "transformer_ff": 4096,
-      "hidden_size": 1024,
+      "add_ffnbias": true,
       "add_qkvbias": false,
       "layer_norm": "standard",
+      "src_word_vec_size": 768,
+      "hidden_size": 768,
+      "layers": 12,
       "n_positions": null,
-      "add_ffnbias": true,
-      "decoder_type": "transformer",
-      "mlp_activation_fn": "gelu",
-      "layers": 2,
-      "norm_eps": 1e-06,
       "position_encoding_type": "SinusoidalInterleaved"
     }
   },
-  "data": {
-    "corpus_1": {
-      "path_src": "hf://quickmt/quickmt-train.ar-en/ar",
-      "path_align": null,
-      "transforms": [
-        "sentencepiece",
-        "filtertoolong"
-      ],
-      "path_tgt": "hf://quickmt/quickmt-train.ar-en/en",
-      "path_sco": "hf://quickmt/quickmt-train.ar-en/sco"
+  "transforms_configs": {
+    "filtertoolong": {
+      "tgt_seq_length": 256,
+      "src_seq_length": 256
     },
-    "valid": {
-      "path_src": "flores-dev.ar",
-      "path_tgt": "flores-dev.en",
-      "path_align": null,
-      "transforms": [
-        "sentencepiece",
-        "filtertoolong"
-      ]
+    "sentencepiece": {
+      "tgt_subword_model": "${MODEL_PATH}/en.spm.model",
+      "src_subword_model": "${MODEL_PATH}/ar.spm.model"
     }
   }
 }

eole-model/en.spm.model

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:acf054421b5ed52bd6aee14564fb8595b9a6b492df669e971d43fba9edb33c16
-size 589684
+oid sha256:8c793ee4a6d01a8b3fcd24d246a1580febf053747e44eefe99606974957cfdaa
+size 806180

eole-model/model.00.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f5529927ea7de80e3e4d2a39880dfdbdbd96801c1d46060e1649af9a3266b383
-size 742170000
+oid sha256:07c8ebd41b7dbef57b13fb30d2124d1345dc574d08331b10e406353dffeebc65
+size 829569112

model.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:4a4210397c03c0ebddb84a0beaa39aa51dcaa5dce07dc04f63ec5e02bbc93911
-size 360843109
+oid sha256:2c4e6f5d86de0d9d6c23c54768d7ab6bd20fb7a36099c987368d6a9f45c32da1
+size 407101843

src.spm.model

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:c7b8bf117088a901628696f60ddd15e047057ccbdcf9b0996ca19e2779a234c7
-size 642429
+oid sha256:1a510ccfeb5cf06584327b9c2482ed0a4111d06348a56e28d165165b6dabe520
+size 901383

tgt.spm.model

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:acf054421b5ed52bd6aee14564fb8595b9a6b492df669e971d43fba9edb33c16
-size 589684
+oid sha256:8c793ee4a6d01a8b3fcd24d246a1580febf053747e44eefe99606974957cfdaa
+size 806180