upgrade modal

server/gpu/modal_deployments/reflector_diarizer.py

@@ -6,12 +6,12 @@ Reflector GPU backend - diarizer
 import os
 
 import modal.gpu
-from modal import Image, Secret, Stub, asgi_app, method
+from modal import Image, Secret, App, asgi_app, method, enter
 from pydantic import BaseModel
 
-PYANNOTE_MODEL_NAME: str = "pyannote/speaker-diarization-3.0"
+PYANNOTE_MODEL_NAME: str = "pyannote/speaker-diarization-3.1"
 MODEL_DIR = "/root/diarization_models"
-stub = Stub(name="reflector-diarizer")
+app = App(name="reflector-diarizer")
 
 
 def migrate_cache_llm():
@@ -33,7 +33,6 @@ def download_pyannote_audio():
     Pipeline.from_pretrained(
         "pyannote/speaker-diarization-3.0",
         cache_dir=MODEL_DIR,
-        use_auth_token=os.environ["HF_TOKEN"]
     )
 
 
@@ -54,7 +53,7 @@ diarizer_image = (
         "hf-transfer"
     )
     .run_function(migrate_cache_llm)
-    .run_function(download_pyannote_audio, secrets=[modal.Secret.from_name("my-huggingface-secret")])
+    .run_function(download_pyannote_audio)
     .env(
         {
             "LD_LIBRARY_PATH": (
@@ -66,16 +65,16 @@ diarizer_image = (
 )
 
 
-@stub.cls(
+@app.cls(
     gpu=modal.gpu.A100(memory=40),
     timeout=60 * 30,
     container_idle_timeout=60,
     allow_concurrent_inputs=1,
     image=diarizer_image,
-    secrets=[modal.Secret.from_name("my-huggingface-secret")],
 )
 class Diarizer:
-    def __enter__(self):
+    @enter()
+    def enter(self):
         import torch
         from pyannote.audio import Pipeline
 
@@ -124,7 +123,7 @@ class Diarizer:
 # -------------------------------------------------------------------
 
 
-@stub.function(
+@app.function(
     timeout=60 * 10,
     container_idle_timeout=60 * 3,
     allow_concurrent_inputs=40,

server/gpu/modal_deployments/reflector_llm.py

@@ -9,7 +9,7 @@ import threading
 from typing import Optional
 
 import modal
-from modal import Image, Secret, Stub, asgi_app, method
+from modal import Image, Secret, App, asgi_app, method, enter, exit
 
 # LLM
 LLM_MODEL: str = "lmsys/vicuna-13b-v1.5"
@@ -19,7 +19,7 @@ LLM_MAX_NEW_TOKENS: int = 300
 
 IMAGE_MODEL_DIR = "/root/llm_models"
 
-stub = Stub(name="reflector-llm")
+app = App(name="reflector-llm")
 
 
 def download_llm():
@@ -64,7 +64,7 @@ llm_image = (
 )
 
 
-@stub.cls(
+@app.cls(
     gpu="A100",
     timeout=60 * 5,
     container_idle_timeout=60 * 5,
@@ -72,7 +72,8 @@ llm_image = (
     image=llm_image,
 )
 class LLM:
-    def __enter__(self):
+    @enter()
+    def enter(self):
         import torch
         from transformers import AutoModelForCausalLM, AutoTokenizer, GenerationConfig
 
@@ -113,7 +114,8 @@ class LLM:
 
         self.lock = threading.Lock()
 
-    def __exit__(self, *args):
+    @exit()
+    def exit():
         print("Exit llm")
 
     @method()
@@ -161,7 +163,7 @@ class LLM:
 # -------------------------------------------------------------------
 
 
-@stub.function(
+@app.function(
     container_idle_timeout=60 * 10,
     timeout=60 * 5,
     allow_concurrent_inputs=45,

server/gpu/modal_deployments/reflector_llm_zephyr.py

@@ -9,7 +9,7 @@ import threading
 from typing import Optional
 
 import modal
-from modal import Image, Secret, Stub, asgi_app, method
+from modal import Image, Secret, App, asgi_app, method, enter, exit
 
 # LLM
 LLM_MODEL: str = "HuggingFaceH4/zephyr-7b-alpha"
@@ -19,7 +19,7 @@ LLM_MAX_NEW_TOKENS: int = 300
 
 IMAGE_MODEL_DIR = "/root/llm_models/zephyr"
 
-stub = Stub(name="reflector-llm-zephyr")
+app = App(name="reflector-llm-zephyr")
 
 
 def download_llm():
@@ -64,7 +64,7 @@ llm_image = (
 )
 
 
-@stub.cls(
+@app.cls(
     gpu="A10G",
     timeout=60 * 5,
     container_idle_timeout=60 * 5,
@@ -72,7 +72,8 @@ llm_image = (
     image=llm_image,
 )
 class LLM:
-    def __enter__(self):
+    @enter()
+    def enter(self):
         import torch
         from transformers import AutoModelForCausalLM, AutoTokenizer, GenerationConfig
 
@@ -116,7 +117,8 @@ class LLM:
         self.GenerationConfig = GenerationConfig
         self.lock = threading.Lock()
 
-    def __exit__(self, *args):
+    @exit()
+    def exit():
         print("Exit llm")
 
     @method()
@@ -169,7 +171,7 @@ class LLM:
 # -------------------------------------------------------------------
 
 
-@stub.function(
+@app.function(
     container_idle_timeout=60 * 10,
     timeout=60 * 5,
     allow_concurrent_inputs=30,

server/gpu/modal_deployments/reflector_transcriber.py

@@ -7,7 +7,7 @@ import os
 import tempfile
 import threading
 
-from modal import Image, Secret, Stub, asgi_app, method
+from modal import Image, Secret, App, asgi_app, method, enter
 from pydantic import BaseModel
 
 # Whisper
@@ -18,7 +18,7 @@ WHISPER_NUM_WORKERS: int = 1
 
 WHISPER_MODEL_DIR = "/root/transcription_models"
 
-stub = Stub(name="reflector-transcriber")
+app = App(name="reflector-transcriber")
 
 
 def download_whisper():
@@ -75,7 +75,7 @@ transcriber_image = (
 )
 
 
-@stub.cls(
+@app.cls(
     gpu="A10G",
     timeout=60 * 5,
     container_idle_timeout=60 * 5,
@@ -83,7 +83,8 @@ transcriber_image = (
     image=transcriber_image,
 )
 class Transcriber:
-    def __enter__(self):
+    @enter()
+    def enter(self):
         import faster_whisper
         import torch
 
@@ -149,7 +150,7 @@ class Transcriber:
 # -------------------------------------------------------------------
 
 
-@stub.function(
+@app.function(
     container_idle_timeout=60,
     timeout=60,
     allow_concurrent_inputs=40,

server/gpu/modal_deployments/reflector_translator.py

@@ -6,7 +6,7 @@ Reflector GPU backend - transcriber
 import os
 import threading
 
-from modal import Image, Secret, Stub, asgi_app, method
+from modal import Image, Secret, App, asgi_app, method, enter
 from pydantic import BaseModel
 
 # Seamless M4T
@@ -20,7 +20,7 @@ HF_SEAMLESS_M4T_VOCODEREPO: str = "facebook/seamless-m4t-vocoder"
 SEAMLESS_GITEPO: str = "https://github.com/facebookresearch/seamless_communication.git"
 SEAMLESS_MODEL_DIR: str = "m4t"
 
-stub = Stub(name="reflector-translator")
+app = App(name="reflector-translator")
 
 
 def install_seamless_communication():
@@ -134,7 +134,7 @@ transcriber_image = (
 )
 
 
-@stub.cls(
+@app.cls(
     gpu="A10G",
     timeout=60 * 5,
     container_idle_timeout=60 * 5,
@@ -142,7 +142,8 @@ transcriber_image = (
     image=transcriber_image,
 )
 class Translator:
-    def __enter__(self):
+    @enter()
+    def enter(self):
         import torch
         from seamless_communication.inference.translator import Translator
 
@@ -379,7 +380,7 @@ class Translator:
 # -------------------------------------------------------------------
 
 
-@stub.function(
+@app.function(
     container_idle_timeout=60,
     timeout=60,
     allow_concurrent_inputs=40,