Initial Commit to Test Runpod

Dockerfile

@@ -0,0 +1,16 @@
+FROM nvidia/cuda:12.1.1-cudnn8-devel-ubuntu22.04
+ENV DEBIAN_FRONTEND=noninteractive
+
+RUN useradd -m -u 1000 user
+USER user
+ENV HOME=/home/user \
+    PATH=/home/user/.local/bin:${PATH}
+WORKDIR ${HOME}/app
+
+COPY --chown=1000 . ${HOME}/app
+RUN pip install -r ${HOME}/app/requirements.txt && \
+    pip install fairseq2 --pre --extra-index-url https://fair.pkg.atmeta.com/fairseq2/pt2.1.0/cu121 && \
+    pip install ${HOME}/app/whl/seamless_communication-1.0.0-py3-none-any.whl
+# This will cache the model into the docker image
+RUN python -u translator.py
+CMD ["python", "main.py"]

lang_list.py

@@ -0,0 +1,255 @@
+# Language dict
+language_code_to_name = {
+    "afr": "Afrikaans",
+    "amh": "Amharic",
+    "arb": "Modern Standard Arabic",
+    "ary": "Moroccan Arabic",
+    "arz": "Egyptian Arabic",
+    "asm": "Assamese",
+    "ast": "Asturian",
+    "azj": "North Azerbaijani",
+    "bel": "Belarusian",
+    "ben": "Bengali",
+    "bos": "Bosnian",
+    "bul": "Bulgarian",
+    "cat": "Catalan",
+    "ceb": "Cebuano",
+    "ces": "Czech",
+    "ckb": "Central Kurdish",
+    "cmn": "Mandarin Chinese",
+    "cym": "Welsh",
+    "dan": "Danish",
+    "deu": "German",
+    "ell": "Greek",
+    "eng": "English",
+    "est": "Estonian",
+    "eus": "Basque",
+    "fin": "Finnish",
+    "fra": "French",
+    "gaz": "West Central Oromo",
+    "gle": "Irish",
+    "glg": "Galician",
+    "guj": "Gujarati",
+    "heb": "Hebrew",
+    "hin": "Hindi",
+    "hrv": "Croatian",
+    "hun": "Hungarian",
+    "hye": "Armenian",
+    "ibo": "Igbo",
+    "ind": "Indonesian",
+    "isl": "Icelandic",
+    "ita": "Italian",
+    "jav": "Javanese",
+    "jpn": "Japanese",
+    "kam": "Kamba",
+    "kan": "Kannada",
+    "kat": "Georgian",
+    "kaz": "Kazakh",
+    "kea": "Kabuverdianu",
+    "khk": "Halh Mongolian",
+    "khm": "Khmer",
+    "kir": "Kyrgyz",
+    "kor": "Korean",
+    "lao": "Lao",
+    "lit": "Lithuanian",
+    "ltz": "Luxembourgish",
+    "lug": "Ganda",
+    "luo": "Luo",
+    "lvs": "Standard Latvian",
+    "mai": "Maithili",
+    "mal": "Malayalam",
+    "mar": "Marathi",
+    "mkd": "Macedonian",
+    "mlt": "Maltese",
+    "mni": "Meitei",
+    "mya": "Burmese",
+    "nld": "Dutch",
+    "nno": "Norwegian Nynorsk",
+    "nob": "Norwegian Bokm\u00e5l",
+    "npi": "Nepali",
+    "nya": "Nyanja",
+    "oci": "Occitan",
+    "ory": "Odia",
+    "pan": "Punjabi",
+    "pbt": "Southern Pashto",
+    "pes": "Western Persian",
+    "pol": "Polish",
+    "por": "Portuguese",
+    "ron": "Romanian",
+    "rus": "Russian",
+    "slk": "Slovak",
+    "slv": "Slovenian",
+    "sna": "Shona",
+    "snd": "Sindhi",
+    "som": "Somali",
+    "spa": "Spanish",
+    "srp": "Serbian",
+    "swe": "Swedish",
+    "swh": "Swahili",
+    "tam": "Tamil",
+    "tel": "Telugu",
+    "tgk": "Tajik",
+    "tgl": "Tagalog",
+    "tha": "Thai",
+    "tur": "Turkish",
+    "ukr": "Ukrainian",
+    "urd": "Urdu",
+    "uzn": "Northern Uzbek",
+    "vie": "Vietnamese",
+    "xho": "Xhosa",
+    "yor": "Yoruba",
+    "yue": "Cantonese",
+    "zlm": "Colloquial Malay",
+    "zsm": "Standard Malay",
+    "zul": "Zulu",
+}
+LANGUAGE_NAME_TO_CODE = {v: k for k, v in language_code_to_name.items()}
+
+# Source langs: S2ST / S2TT / ASR don't need source lang
+# T2TT / T2ST use this
+text_source_language_codes = [
+    "afr",
+    "amh",
+    "arb",
+    "ary",
+    "arz",
+    "asm",
+    "azj",
+    "bel",
+    "ben",
+    "bos",
+    "bul",
+    "cat",
+    "ceb",
+    "ces",
+    "ckb",
+    "cmn",
+    "cym",
+    "dan",
+    "deu",
+    "ell",
+    "eng",
+    "est",
+    "eus",
+    "fin",
+    "fra",
+    "gaz",
+    "gle",
+    "glg",
+    "guj",
+    "heb",
+    "hin",
+    "hrv",
+    "hun",
+    "hye",
+    "ibo",
+    "ind",
+    "isl",
+    "ita",
+    "jav",
+    "jpn",
+    "kan",
+    "kat",
+    "kaz",
+    "khk",
+    "khm",
+    "kir",
+    "kor",
+    "lao",
+    "lit",
+    "lug",
+    "luo",
+    "lvs",
+    "mai",
+    "mal",
+    "mar",
+    "mkd",
+    "mlt",
+    "mni",
+    "mya",
+    "nld",
+    "nno",
+    "nob",
+    "npi",
+    "nya",
+    "ory",
+    "pan",
+    "pbt",
+    "pes",
+    "pol",
+    "por",
+    "ron",
+    "rus",
+    "slk",
+    "slv",
+    "sna",
+    "snd",
+    "som",
+    "spa",
+    "srp",
+    "swe",
+    "swh",
+    "tam",
+    "tel",
+    "tgk",
+    "tgl",
+    "tha",
+    "tur",
+    "ukr",
+    "urd",
+    "uzn",
+    "vie",
+    "yor",
+    "yue",
+    "zsm",
+    "zul",
+]
+TEXT_SOURCE_LANGUAGE_NAMES = sorted([language_code_to_name[code] for code in text_source_language_codes])
+
+# Target langs:
+# S2ST / T2ST
+s2st_target_language_codes = [
+    "eng",
+    "arb",
+    "ben",
+    "cat",
+    "ces",
+    "cmn",
+    "cym",
+    "dan",
+    "deu",
+    "est",
+    "fin",
+    "fra",
+    "hin",
+    "ind",
+    "ita",
+    "jpn",
+    "kor",
+    "mlt",
+    "nld",
+    "pes",
+    "pol",
+    "por",
+    "ron",
+    "rus",
+    "slk",
+    "spa",
+    "swe",
+    "swh",
+    "tel",
+    "tgl",
+    "tha",
+    "tur",
+    "ukr",
+    "urd",
+    "uzn",
+    "vie",
+]
+S2ST_TARGET_LANGUAGE_NAMES = sorted([language_code_to_name[code] for code in s2st_target_language_codes])
+T2ST_TARGET_LANGUAGE_NAMES = S2ST_TARGET_LANGUAGE_NAMES
+
+# S2TT / T2TT / ASR
+S2TT_TARGET_LANGUAGE_NAMES = TEXT_SOURCE_LANGUAGE_NAMES
+T2TT_TARGET_LANGUAGE_NAMES = TEXT_SOURCE_LANGUAGE_NAMES
+ASR_TARGET_LANGUAGE_NAMES = TEXT_SOURCE_LANGUAGE_NAMES

main.py

@@ -0,0 +1,25 @@
+# from translator import translator
+from lang_list import LANGUAGE_NAME_TO_CODE
+import runpod
+
+def run_t2tt(input_text: str, source_language: str, target_language: str) -> str:
+    source_language_code = LANGUAGE_NAME_TO_CODE[source_language]
+    target_language_code = LANGUAGE_NAME_TO_CODE[target_language]
+    # out_texts, _ = translator.predict(
+    #     input=input_text,
+    #     task_str="T2TT",
+    #     src_lang=source_language_code,
+    #     tgt_lang=target_language_code,
+    # )
+    # return str(out_texts[0])
+    import json
+    return json.dumps({"input_text": input_text, "src_code": source_language_code, "tgt_code": target_language_code})
+
+def runpod_handler(job):
+    job_input = job['input']
+    input_text = job_input["input_text"]
+    source_language = job_input["source_language"]
+    target_language = job_input["target_language"]
+    return run_t2tt(input_text, source_language, target_language)
+
+runpod.serverless.start({"handler": runpod_handler})

requirements.txt

@@ -0,0 +1,5 @@
+gradio==4.9.0
+omegaconf==2.3.0
+torch==2.1.0
+torchaudio==2.1.0
+runpod

test_input.json

@@ -0,0 +1,7 @@
+{
+    "input": {
+        "input_text": "How are you doing today?",
+        "source_language": "English",
+        "target_language": "Mandarin Chinese"
+    }
+}

translator.py

@@ -0,0 +1,39 @@
+import os
+import pathlib
+import torch
+from fairseq2.assets import InProcAssetMetadataProvider, asset_store
+from huggingface_hub import snapshot_download
+from seamless_communication.inference import Translator
+
+CHECKPOINTS_PATH = pathlib.Path(os.getenv("CHECKPOINTS_PATH", "/home/user/app/models"))
+if not CHECKPOINTS_PATH.exists():
+    snapshot_download(repo_id="facebook/seamless-m4t-v2-large", repo_type="model", local_dir=CHECKPOINTS_PATH)
+asset_store.env_resolvers.clear()
+asset_store.env_resolvers.append(lambda: "demo")
+demo_metadata = [
+    {
+        "name": "seamlessM4T_v2_large@demo",
+        "checkpoint": f"file://{CHECKPOINTS_PATH}/seamlessM4T_v2_large.pt",
+        "char_tokenizer": f"file://{CHECKPOINTS_PATH}/spm_char_lang38_tc.model",
+    },
+    {
+        "name": "vocoder_v2@demo",
+        "checkpoint": f"file://{CHECKPOINTS_PATH}/vocoder_v2.pt",
+    },
+]
+asset_store.metadata_providers.append(InProcAssetMetadataProvider(demo_metadata))
+
+if torch.cuda.is_available():
+    device = torch.device("cuda:0")
+    dtype = torch.float16
+else:
+    device = torch.device("cpu")
+    dtype = torch.float32
+
+translator = Translator(
+    model_name_or_card="seamlessM4T_v2_large",
+    vocoder_name_or_card="vocoder_v2",
+    device=device,
+    dtype=dtype,
+    apply_mintox=True,
+)