Adjust setup.py. Map scripts to binary names. Adjust Readme docs.

README.md

@@ -22,22 +22,27 @@ Links:
 
 # Quick Start
 ## Installation
+
 ```
-pip install fairseq2==0.1
 pip install .
 ```
 
+A temporary extra requirement for fairseq2 is [libsndfile](https://github.com/libsndfile/libsndfile). From [Conda](https://docs.conda.io/en/latest/) environment it can be installed via:
+```
+conda install -y -c conda-forge libsndfile
+```
+
 ## Running inference
 
 Here’s an example of using the CLI from the root directory to run inference.
 
 S2ST task:
 ```bash
-python scripts/m4t/predict/predict.py <path_to_input_audio> s2st <tgt_lang> --output_path <path_to_save_audio>
+m4t_predict <path_to_input_audio> s2st <tgt_lang> --output_path <path_to_save_audio>
 ```
 T2TT task:
 ```bash
-python scripts/m4t/predict/predict.py <input_text> t2tt <tgt_lang> --src_lang <src_lang>
+m4t_predict <input_text> t2tt <tgt_lang> --src_lang <src_lang>
 ```
 
 Please refer to the [evaluation README](scripts/m4t/predict) for detailed instruction on how to run inference.

dev_requirements.txt

@@ -0,0 +1,4 @@
+pytest
+black
+flake8
+isort

requirements.txt

@@ -3,3 +3,4 @@ datasets
 torchaudio
 soundfile
 librosa
+fairseq2

scripts/m4t/finetune/README.md

@@ -29,12 +29,12 @@ Below is an example bash script that prepares a training and evaluation dataset
 export DATASET_DIR=~/m4t_dataset
 mkdir -p $DATASET_DIR
 
-python scripts/m4t/finetune/dataset.py \
+m4t_prepare_dataset \
   --source_lang eng \
   --target_lang kor \
   --split train \
   --save_dir $DATASET_DIR
- python scripts/m4t/finetune/dataset.py \
+m4t_prepare_dataset \
   --source_lang eng \
   --target_lang kor \
   --split validation \
@@ -97,7 +97,8 @@ torchrun \
    --rdzv-endpoint=localhost:0 \
    --nnodes=1 \
    --nproc-per-node=8  \
-  scripts/m4t/finetune/finetune.py \
+   --no-python \
+  m4t_finetune \
    --mode SPEECH_TO_TEXT \
    --train_dataset $DATASET_DIR/train_manifest.json  \
    --eval_dataset $DATASET_DIR/validation_manifest.json \

scripts/m4t/finetune/dataset.py

@@ -10,7 +10,6 @@ import dataclasses
 import json
 import logging
 import os
-from argparse import Namespace
 from pathlib import Path
 
 from seamless_communication.datasets.huggingface import (
@@ -157,7 +156,8 @@ def init_parser() -> argparse.ArgumentParser:
     return parser
 
 
-def main(args: Namespace) -> None:
+def main() -> None:
+    args = init_parser().parse_args()
     manifest_path = download_fleurs_dataset(
         source_lang=args.source_lang,
         target_lang=args.target_lang,
@@ -168,5 +168,4 @@ def main(args: Namespace) -> None:
 
 
 if __name__ == "__main__":
-    args = init_parser().parse_args()
-    main(args)
+    main()

scripts/m4t/finetune/finetune.py

@@ -7,7 +7,6 @@
 import argparse
 import logging
 import os
-from argparse import Namespace
 from pathlib import Path
 
 import dataloader
@@ -125,7 +124,8 @@ def init_parser() -> argparse.ArgumentParser:
     return parser
 
 
-def run_finetune(args: Namespace) -> None:
+def main() -> None:
+    args = init_parser().parse_args()
     dist_utils.init_distributed([logger, trainer.logger])
     device = torch.device("cuda")
     text_tokenizer: NllbTokenizer = load_unity_text_tokenizer(args.model_name)
@@ -182,5 +182,4 @@ def run_finetune(args: Namespace) -> None:
 
 
 if __name__ == "__main__":
-    parser = init_parser()
-    run_finetune(parser.parse_args())
+    main()

scripts/m4t/predict/predict.py

@@ -9,8 +9,11 @@ import torch
 import torchaudio
 from seamless_communication.models.inference import Translator
 
+logging.basicConfig(
+    level=logging.INFO,
+    format="%(asctime)s %(levelname)s -- %(name)s: %(message)s",
+)
 
-logging.basicConfig(level=logging.INFO)
 logger = logging.getLogger(__name__)
 
 

setup.py

@@ -4,12 +4,44 @@
 # This source code is licensed under the license found in the
 # LICENSE file in the root directory of this source tree.
 
+from pathlib import Path
+from typing import Iterable
+
+import pkg_resources
 from setuptools import find_packages, setup
 
+
+def _load_requirements(fname: str) -> Iterable[str]:
+    with open(Path(__file__).parent / fname) as fp_in:
+        for req in pkg_resources.parse_requirements(fp_in):
+            yield str(req)
+
+
+default_requirements = list(_load_requirements("requirements.txt"))
+dev_requirements = list(_load_requirements("dev_requirements.txt"))
+
 setup(
     name="seamless_communication",
-    version="0.1",
-    packages=find_packages(where="src"),
-    package_dir={"": "src"},
-    package_data={"": ["assets/cards/*.yaml"]},
+    version="1.0.0",
+    packages=find_packages(where="src") + ['m4t_scripts.finetune', 'm4t_scripts.predict'],
+    package_dir={"m4t_scripts": "scripts/m4t", "seamless_communication": "src/seamless_communication"},
+    package_data={"": ["seamless_communication/assets/cards/*.yaml"]},
+    description="SeamlessM4T -- Massively Multilingual & Multimodal Machine Translation Model",
+    long_description=open("README.md", encoding="utf-8").read(),
+    long_description_content_type="text/markdown",
+    readme="README.md",
+    python_requires=">=3.8",
+    author="Meta Platforms",
+    url="https://ai.meta.com/",
+    license="Creative Commons",
+    install_requires=default_requirements,
+    extras_require={"dev": default_requirements + dev_requirements},
+    entry_points={
+        "console_scripts": [
+           "m4t_predict=m4t_scripts.predict.predict:main",
+           "m4t_finetune=m4t_scripts.finetune.finetune:main",
+           "m4t_prepare_dataset=m4t_scripts.finetune.dataset:main",
+        ],
+    },
+    include_package_data=True,
 )