Add modelscope to all language libraries

node-hub/dora-distil-whisper/dora_distil_whisper/main.py

@@ -3,11 +3,21 @@ from transformers import AutoModelForSpeechSeq2Seq, AutoProcessor, pipeline
 from dora import Node
 import pyarrow as pa
 import os
+from pathlib import Path
 
-MODEL_NAME_OR_PATH = os.getenv("MODEL_NAME_OR_PATH", "openai/whisper-large-v3-turbo")
+DEFAULT_PATH = "openai/whisper-large-v3-turbo"
 TARGET_LANGUAGE = os.getenv("TARGET_LANGUAGE", "chinese")
 TRANSLATE = bool(os.getenv("TRANSLATE", "False"))
 
+
+MODEL_NAME_OR_PATH = os.getenv("MODEL_NAME_OR_PATH", DEFAULT_PATH)
+
+if bool(os.getenv("USE_MODELSCOPE_HUB")) is True:
+    from modelscope import snapshot_download
+
+    if not Path(MODEL_NAME_OR_PATH).exists():
+        MODEL_NAME_OR_PATH = snapshot_download(MODEL_NAME_OR_PATH)
+
 device = "cuda:0" if torch.cuda.is_available() else "cpu"
 torch_dtype = torch.float16 if torch.cuda.is_available() else torch.float32
 

node-hub/dora-distil-whisper/pyproject.toml

@@ -20,6 +20,7 @@ transformers = "^4.0.0"
 accelerate = "^0.29.2"
 torch = "^2.2.0"
 python = "^3.7"
+modelscope = "^1.18.1"
 
 [tool.poetry.scripts]
 dora-distil-whisper = "dora_distil_whisper.main:main"

node-hub/dora-opus/dora_opus/main.py

@@ -1,15 +1,26 @@
 import os
+from pathlib import Path
 from dora import Node
 import pyarrow as pa
 import numpy as np
+from transformers import AutoTokenizer, AutoModelForSeq2SeqLM
 
 from_code = os.getenv("SOURCE_LANGUAGE", "zh")
 to_code = os.getenv("TARGET_LANGUAGE", "en")
-MODEL_NAME_OR_PATH = os.getenv(
-    "MODEL_NAME_OR_PATH", f"Helsinki-NLP/opus-mt-{from_code}-{to_code}"
-)
+DEFAULT_PATH = f"Helsinki-NLP/opus-mt-{from_code}-{to_code}"
 
-from transformers import AutoTokenizer, AutoModelForSeq2SeqLM
+
+MODEL_NAME_OR_PATH = os.getenv("MODEL_NAME_OR_PATH", DEFAULT_PATH)
+
+if bool(os.getenv("USE_MODELSCOPE_HUB")) is True:
+    from modelscope import snapshot_download
+
+    if not Path(MODEL_NAME_OR_PATH).exists():
+        MODEL_NAME_OR_PATH = snapshot_download(MODEL_NAME_OR_PATH)
+
+tokenizer = AutoTokenizer.from_pretrained(MODEL_NAME_OR_PATH)
+
+model = AutoModelForSeq2SeqLM.from_pretrained(MODEL_NAME_OR_PATH)
 
 
 def cut_repetition(text, min_repeat_length=4, max_repeat_length=50):
@@ -42,9 +53,6 @@ def cut_repetition(text, min_repeat_length=4, max_repeat_length=50):
 
 
 def main():
-    tokenizer = AutoTokenizer.from_pretrained(MODEL_NAME_OR_PATH)
-
-    model = AutoModelForSeq2SeqLM.from_pretrained(MODEL_NAME_OR_PATH)
     node = Node()
     while True:
         event = node.next()

node-hub/dora-opus/pyproject.toml

@@ -16,6 +16,7 @@ dora-rs = "^0.3.6"
 numpy = "< 2.0.0"
 python = "^3.7"
 transformers = "^4.45"
+modelscope = "^1.18.1"
 
 [tool.poetry.scripts]
 dora-opus = "dora_opus.main:main"

node-hub/dora-parler/dora_parler/main.py

@@ -1,6 +1,7 @@
 from threading import Thread
 from dora import Node
-
+import os
+from pathlib import Path
 import numpy as np
 import torch
 import time
@@ -18,16 +19,25 @@ from transformers import (
 device = "cuda:0"  # if torch.cuda.is_available() else "mps" if torch.backends.mps.is_available() else "cpu"
 torch_dtype = torch.float16 if device != "cpu" else torch.float32
 
-repo_id = "ylacombe/parler-tts-mini-jenny-30H"
+DEFAULT_PATH = "ylacombe/parler-tts-mini-jenny-30H"
+
+
+MODEL_NAME_OR_PATH = os.getenv("MODEL_NAME_OR_PATH", DEFAULT_PATH)
+
+if bool(os.getenv("USE_MODELSCOPE_HUB")) is True:
+    from modelscope import snapshot_download
+
+    if not Path(MODEL_NAME_OR_PATH).exists():
+        MODEL_NAME_OR_PATH = snapshot_download(MODEL_NAME_OR_PATH)
 
 model = ParlerTTSForConditionalGeneration.from_pretrained(
-    repo_id, torch_dtype=torch_dtype, low_cpu_mem_usage=True
+    MODEL_NAME_OR_PATH, torch_dtype=torch_dtype, low_cpu_mem_usage=True
 ).to(device)
 model.generation_config.cache_implementation = "static"
 model.forward = torch.compile(model.forward, mode="default")
 
-tokenizer = AutoTokenizer.from_pretrained(repo_id)
-feature_extractor = AutoFeatureExtractor.from_pretrained(repo_id)
+tokenizer = AutoTokenizer.from_pretrained(MODEL_NAME_OR_PATH)
+feature_extractor = AutoFeatureExtractor.from_pretrained(MODEL_NAME_OR_PATH)
 
 SAMPLE_RATE = feature_extractor.sampling_rate
 SEED = 42
@@ -59,6 +69,7 @@ def play_audio(audio_array):
 
 class InterruptStoppingCriteria(StoppingCriteria):
     def __init__(self):
+        super().__init__()
         self.stop_signal = False
 
     def __call__(

node-hub/dora-parler/pyproject.toml

@@ -20,6 +20,7 @@ torchaudio = "^2.2.2"
 sentencepiece = "^0.1.99"
 python = "^3.7"
 pyaudio = "^0.2.14"
+modelscope = "^1.18.1"
 
 
 [tool.poetry.scripts]

node-hub/dora-qwenvl/dora_qwenvl/main.py

@@ -12,7 +12,7 @@ DEFAULT_PATH = "Qwen/Qwen2-VL-2B-Instruct"
 
 MODEL_NAME_OR_PATH = os.getenv("MODEL_NAME_OR_PATH", DEFAULT_PATH)
 
-if bool(os.getenv("MODELSCOPE")) is True:
+if bool(os.getenv("USE_MODELSCOPE_HUB")) is True:
     from modelscope import snapshot_download
 
     if not Path(MODEL_NAME_OR_PATH).exists():