Upgrade AllenNLP + hotfix (#287)

* Fix AllenNLP inference

* Fix API GA workflow

* Fix dataset name of batch test

* Fix the docker GA workflow

* Ensure AllenNLP test is actually running and not just skipped

* Change env variable

* Fix style, upgrade allennlp, revert some changes

* Fix dependencies of AllenNLP

* Test longer retry

* Format

* Test again

Author: osanseviero

.github/workflows/python-api-allennlp.yaml

@@ -5,13 +5,15 @@ on:
     paths:
       - "api-inference-community/docker_images/allennlp/**"
 jobs:
-  docker:
+  build:
     runs-on: ubuntu-latest
     steps:
       - name: Set up Python ${{ matrix.python-version }}
         uses: actions/setup-python@v2
         with:
           python-version: "3.8"
+      - name: Checkout
+        uses: actions/checkout@v2
       - name: Set up QEMU
         uses: docker/setup-qemu-action@v1
       - name: Set up Docker Buildx
@@ -22,5 +24,5 @@ jobs:
           pip install --upgrade pip
           pip install pytest pillow httpx
           pip install -e .
-      - run: RUN_DOCKER=1 pytest -sv tests/test_dockers.py::DockerImageTests::test_allennlp
+      - run: RUN_DOCKER_TESTS=1 pytest -sv tests/test_dockers.py::DockerImageTests::test_allennlp
         working-directory: api-inference-community

.github/workflows/python-api-tests.yaml

@@ -12,6 +12,7 @@ jobs:
 
     steps:
       - run: |
+          sudo apt-get update
           sudo apt-get install ffmpeg
 
       - uses: actions/checkout@v2

api-inference-community/Makefile

@@ -1,7 +1,7 @@
 .PHONY: quality style
 
 
-check_dirs := api_inference_community tests
+check_dirs := api_inference_community tests docker_images
 
 
 

api-inference-community/docker_images/allennlp/app/main.py

@@ -1,13 +1,20 @@
+import functools
 import logging
 import os
 from typing import Dict, Type
 
+from api_inference_community.routes import pipeline_route, status_ok
 from app.pipelines import Pipeline, QuestionAnsweringPipeline
-from app.routes import pipeline_route, status_ok
 from starlette.applications import Starlette
+from starlette.middleware import Middleware
+from starlette.middleware.gzip import GZipMiddleware
 from starlette.routing import Route
 
 
+TASK = os.getenv("TASK")
+MODEL_ID = os.getenv("MODEL_ID")
+
+
 logger = logging.getLogger(__name__)
 
 
@@ -30,7 +37,10 @@
 }
 
 
-def get_pipeline(task: str, model_id: str) -> Pipeline:
+@functools.lru_cache()
+def get_pipeline() -> Pipeline:
+    task = os.environ["TASK"]
+    model_id = os.environ["MODEL_ID"]
     if task not in ALLOWED_TASKS:
         raise EnvironmentError(f"{task} is not a valid pipeline for model : {model_id}")
     return ALLOWED_TASKS[task](model_id)
@@ -41,14 +51,21 @@ def get_pipeline(task: str, model_id: str) -> Pipeline:
     Route("/{whatever:path}", pipeline_route, methods=["POST"]),
 ]
 
-app = Starlette(routes=routes)
+middleware = [Middleware(GZipMiddleware, minimum_size=1000)]
 if os.environ.get("DEBUG", "") == "1":
     from starlette.middleware.cors import CORSMiddleware
 
-    app.add_middleware(
-        CORSMiddleware, allow_origins=["*"], allow_headers=["*"], allow_methods=["*"]
+    middleware.append(
+        Middleware(
+            CORSMiddleware,
+            allow_origins=["*"],
+            allow_headers=["*"],
+            allow_methods=["*"],
+        )
     )
 
+app = Starlette(routes=routes, middleware=middleware)
+
 
 @app.on_event("startup")
 async def startup_event():
@@ -57,14 +74,18 @@ async def startup_event():
     handler.setFormatter(logging.Formatter("%(asctime)s - %(levelname)s - %(message)s"))
     logger.handlers = [handler]
 
-    task = os.environ["TASK"]
-    model_id = os.environ["MODEL_ID"]
-
-    app.pipeline = get_pipeline(task, model_id)
+    # Link between `api-inference-community` and framework code.
+    app.get_pipeline = get_pipeline
+    try:
+        get_pipeline()
+    except Exception:
+        # We can fail so we can show exception later.
+        pass
 
 
 if __name__ == "__main__":
-    task = os.environ["TASK"]
-    model_id = os.environ["MODEL_ID"]
-
-    get_pipeline(task, model_id)
+    try:
+        get_pipeline()
+    except Exception:
+        # We can fail so we can show exception later.
+        pass

api-inference-community/docker_images/allennlp/app/pipelines/question_answering.py

@@ -11,7 +11,7 @@ def __init__(
         self,
         model_id: str,
     ):
-        self.predictor = Predictor.from_path(model_id)
+        self.predictor = Predictor.from_path("hf://" + model_id)
 
     def __call__(self, inputs: Dict[str, str]) -> Dict[str, Any]:
         """