trigger full

.github/workflows/doctest_job.yml

@@ -0,0 +1,81 @@
+name: Doctest job
+
+on:
+  workflow_call:
+    inputs:
+      job_splits:
+        required: true
+        type: string
+      split_keys:
+        required: true
+        type: string
+
+env:
+  HF_HOME: /mnt/cache
+  TRANSFORMERS_IS_CI: yes
+  RUN_SLOW: yes
+  OMP_NUM_THREADS: 16
+  MKL_NUM_THREADS: 16
+  SIGOPT_API_TOKEN: ${{ secrets.SIGOPT_API_TOKEN }}
+  TF_FORCE_GPU_ALLOW_GROWTH: true
+
+jobs:
+  run_doctests:
+    name: " "
+    strategy:
+      fail-fast: false
+      matrix:
+        split_keys: ${{ fromJson(inputs.split_keys) }}
+    runs-on: [single-gpu, nvidia-gpu, t4, ci]
+    container:
+      image: huggingface/transformers-all-latest-gpu
+      options: --gpus 0 --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
+    steps:
+      - name: Update clone
+        working-directory: /transformers
+        run: git fetch && git checkout ${{ github.sha }}
+
+      - name: Reinstall transformers in edit mode (remove the one installed during docker image build)
+        working-directory: /transformers
+        run: python3 -m pip uninstall -y transformers && python3 -m pip install -e .[flax]
+
+      - name: GPU visibility
+        working-directory: /transformers
+        run: |
+          python3 utils/print_env.py
+
+      - name: Show installed libraries and their versions
+        run: pip freeze
+
+      - name: Get doctest files
+        working-directory: /transformers
+        run: |
+          echo "${{ toJson(fromJson(inputs.job_splits)[matrix.split_keys]) }}" > doc_tests.txt
+          cat doc_tests.txt
+
+      - name: Set `split_keys`
+        shell: bash
+        run: |
+          echo "${{ matrix.split_keys }}"
+          split_keys=${{ matrix.split_keys }}
+          split_keys=${split_keys//'/'/'_'}
+          echo "split_keys"
+          echo "split_keys=$split_keys" >> $GITHUB_ENV
+
+      - name: Run doctests
+        working-directory: /transformers
+        run: |
+          cat doc_tests.txt
+          python3 -m pytest -v --make-reports doc_tests_gpu_${{ env.split_keys }} --doctest-modules $(cat doc_tests.txt) -sv --doctest-continue-on-failure --doctest-glob="*.md"
+
+      - name: Failure short reports
+        if: ${{ failure() }}
+        continue-on-error: true
+        run: cat /transformers/reports/doc_tests_gpu_${{ env.split_keys }}/failures_short.txt
+
+      - name: "Test suite reports artifacts: doc_tests_gpu_test_reports_${{ env.split_keys }}"
+        if: ${{ always() }}
+        uses: actions/upload-artifact@v3
+        with:
+          name: doc_tests_gpu_test_reports_${{ env.split_keys }}
+          path: /transformers/reports/doc_tests_gpu_${{ env.split_keys }}

.github/workflows/doctests.yml

@@ -3,81 +3,85 @@ name: Doctests
 on:
   push:
     branches:
-      - doctest*
+      - refactor_doctest_2
   repository_dispatch:
   schedule:
     - cron: "17 2 * * *"
 
-
 env:
-  HF_HOME: /mnt/cache
-  TRANSFORMERS_IS_CI: yes
-  RUN_SLOW: yes
-  OMP_NUM_THREADS: 16
-  MKL_NUM_THREADS: 16
-  SIGOPT_API_TOKEN: ${{ secrets.SIGOPT_API_TOKEN }}
-  TF_FORCE_GPU_ALLOW_GROWTH: true
+  NUM_SLICES: 3
 
 jobs:
-  run_doctests:
+  setup:
+    name: Setup
     runs-on: [single-gpu, nvidia-gpu, t4, ci]
     container:
       image: huggingface/transformers-all-latest-gpu
       options: --gpus 0 --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
+    outputs:
+      job_splits: ${{ steps.set-matrix.outputs.job_splits }}
+      split_keys: ${{ steps.set-matrix.outputs.split_keys }}
     steps:
-      - name: uninstall transformers (installed during docker image build)
-        run: python3 -m pip uninstall -y transformers
-
-      - uses: actions/checkout@v3
-      - name: NVIDIA-SMI
+      - name: Update clone
+        working-directory: /transformers
         run: |
-          nvidia-smi
+          git fetch && git checkout ${{ github.sha }}
 
-      - name: Install transformers in edit mode
-        run: python3 -m pip install -e .[flax]
-
-      - name: GPU visibility
-        run: |
-          python3 utils/print_env.py
+      - name: Reinstall transformers in edit mode (remove the one installed during docker image build)
+        working-directory: /transformers
+        run: python3 -m pip uninstall -y transformers && python3 -m pip install -e .
 
       - name: Show installed libraries and their versions
+        working-directory: /transformers
         run: pip freeze
 
-      - name: Get doctest files
+      - name: Check values for matrix
+        working-directory: /transformers
         run: |
-          $(python3 -c 'from utils.tests_fetcher import get_all_doctest_files; to_test = get_all_doctest_files(); to_test = " ".join(to_test); fp = open("doc_tests.txt", "w"); fp.write(to_test); fp.close()')
+          python3 utils/split_doctest_jobs.py
+          python3 utils/split_doctest_jobs.py --only_return_keys --num_splits ${{ env.NUM_SLICES }}
 
-      - name: Run doctests
+      - id: set-matrix
+        working-directory: /transformers
+        name: Set values for matrix
         run: |
-          python3 -m pytest -v --make-reports doc_tests_gpu --doctest-modules $(cat doc_tests.txt) -sv --doctest-continue-on-failure --doctest-glob="*.md"
-
-      - name: Failure short reports
-        if: ${{ failure() }}
-        continue-on-error: true
-        run: cat reports/doc_tests_gpu/failures_short.txt
-
-      - name: Test suite reports artifacts
-        if: ${{ always() }}
-        uses: actions/upload-artifact@v3
-        with:
-          name: doc_tests_gpu_test_reports
-          path: reports/doc_tests_gpu
+          echo "job_splits=$(python3 utils/split_doctest_jobs.py)" >> $GITHUB_OUTPUT
+          echo "split_keys=$(python3 utils/split_doctest_jobs.py --only_return_keys --num_splits ${{ env.NUM_SLICES }})" >> $GITHUB_OUTPUT
 
+  call_doctest_job:
+    name: "Call doctest jobs"
+    needs: setup
+    strategy:
+      fail-fast: false
+      matrix:
+        split_keys: ${{ fromJson(needs.setup.outputs.split_keys) }}
+    uses: ./.github/workflows/doctest_job.yml
+    with:
+      job_splits: ${{ needs.setup.outputs.job_splits }}
+      split_keys: ${{ toJson(matrix.split_keys) }}
+    secrets: inherit
 
   send_results:
     name: Send results to webhook
     runs-on: ubuntu-22.04
     if: always()
-    needs: [run_doctests]
+    needs: [call_doctest_job]
     steps:
       - uses: actions/checkout@v3
       - uses: actions/download-artifact@v3
       - name: Send message to Slack
         env:
           CI_SLACK_BOT_TOKEN: ${{ secrets.CI_SLACK_BOT_TOKEN }}
-          CI_SLACK_CHANNEL_ID: ${{ secrets.CI_SLACK_CHANNEL_ID_DAILY_DOCS }}
-          CI_SLACK_CHANNEL_ID_DAILY: ${{ secrets.CI_SLACK_CHANNEL_ID_DAILY_DOCS }}
-          CI_SLACK_CHANNEL_DUMMY_TESTS: ${{ secrets.CI_SLACK_CHANNEL_DUMMY_TESTS }}
+          ACCESS_REPO_INFO_TOKEN: ${{ secrets.ACCESS_REPO_INFO_TOKEN }}
+          # Use `CI_SLACK_CHANNEL_DUMMY_TESTS` when doing experimentation
+          SLACK_REPORT_CHANNEL: ${{ secrets.CI_SLACK_CHANNEL_ID_DAILY_DOCS }}
         run: |
           pip install slack_sdk
           python utils/notification_service_doc_tests.py
+
+      - name: "Upload results"
+        if: ${{ always() }}
+        uses: actions/upload-artifact@v3
+        with:
+          name: doc_test_results
+          path: doc_test_results

utils/notification_service_doc_tests.py

@@ -12,16 +12,13 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
-import collections
 import json
-import math
 import os
 import re
 import time
-from fnmatch import fnmatch
 from typing import Dict, List
 
-import requests
+from get_ci_error_statistics import get_jobs
 from slack_sdk import WebClient
 
 
@@ -66,9 +63,8 @@ class Message:
     def __init__(self, title: str, doc_test_results: Dict):
         self.title = title
 
-        self._time_spent = doc_test_results["time_spent"].split(",")[0]
-        self.n_success = doc_test_results["success"]
-        self.n_failures = doc_test_results["failures"]
+        self.n_success = sum(job_result["n_success"] for job_result in doc_test_results.values())
+        self.n_failures = sum(job_result["n_failures"] for job_result in doc_test_results.values())
         self.n_tests = self.n_success + self.n_failures
 
         # Failures and success of the modeling tests
@@ -76,7 +72,8 @@ class Message:
 
     @property
     def time(self) -> str:
-        time_spent = [self._time_spent]
+        all_results = [*self.doc_test_results.values()]
+        time_spent = [r["time_spent"].split(", ")[0] for r in all_results if len(r["time_spent"])]
         total_secs = 0
 
         for time in time_spent:
@@ -205,7 +202,7 @@ class Message:
         print(json.dumps({"blocks": json.loads(payload)}))
 
         client.chat_postMessage(
-            channel=os.environ["CI_SLACK_CHANNEL_ID_DAILY"],
+            channel=SLACK_REPORT_CHANNEL_ID,
             text="There was an issue running the tests.",
             blocks=payload,
         )
@@ -217,7 +214,7 @@ class Message:
         text = f"{self.n_failures} failures out of {self.n_tests} tests," if self.n_failures else "All tests passed."
 
         self.thread_ts = client.chat_postMessage(
-            channel=os.environ["CI_SLACK_CHANNEL_ID_DAILY"],
+            channel=SLACK_REPORT_CHANNEL_ID,
             blocks=self.payload,
             text=text,
         )
@@ -248,7 +245,7 @@ class Message:
             }
 
         return [
-            {"type": "header", "text": {"type": "plain_text", "text": title.upper(), "emoji": True}},
+            {"type": "header", "text": {"type": "plain_text", "text": title, "emoji": True}},
             content,
             {"type": "section", "text": {"type": "mrkdwn", "text": failure_text}},
         ]
@@ -257,24 +254,19 @@ class Message:
         if self.thread_ts is None:
             raise ValueError("Can only post reply if a post has been made.")
 
-        job_link = self.doc_test_results.pop("job_link")
-        self.doc_test_results.pop("failures")
-        self.doc_test_results.pop("success")
-        self.doc_test_results.pop("time_spent")
-
         sorted_dict = sorted(self.doc_test_results.items(), key=lambda t: t[0])
-        for job, job_result in sorted_dict:
-            if len(job_result["failures"]):
+        for job_name, job_result in sorted_dict:
+            if len(job_result["failures"]) > 0:
                 text = f"*Num failures* :{len(job_result['failed'])} \n"
                 failures = job_result["failures"]
-                blocks = self.get_reply_blocks(job, job_link, failures, text=text)
+                blocks = self.get_reply_blocks(job_name, job_result["job_link"], failures, text=text)
 
                 print("Sending the following reply")
                 print(json.dumps({"blocks": blocks}))
 
                 client.chat_postMessage(
-                    channel=os.environ["CI_SLACK_CHANNEL_ID_DAILY"],
-                    text=f"Results for {job}",
+                    channel=SLACK_REPORT_CHANNEL_ID,
+                    text=f"Results for {job_name}",
                     blocks=blocks,
                     thread_ts=self.thread_ts["ts"],
                 )
@@ -282,27 +274,6 @@ class Message:
                 time.sleep(1)
 
 
-def get_job_links():
-    run_id = os.environ["GITHUB_RUN_ID"]
-    url = f"https://api.github.com/repos/huggingface/transformers/actions/runs/{run_id}/jobs?per_page=100"
-    result = requests.get(url).json()
-    jobs = {}
-
-    try:
-        jobs.update({job["name"]: job["html_url"] for job in result["jobs"]})
-        pages_to_iterate_over = math.ceil((result["total_count"] - 100) / 100)
-
-        for i in range(pages_to_iterate_over):
-            result = requests.get(url + f"&page={i + 2}").json()
-            jobs.update({job["name"]: job["html_url"] for job in result["jobs"]})
-
-        return jobs
-    except Exception as e:
-        print("Unknown error, could not fetch links.", e)
-
-    return {}
-
-
 def retrieve_artifact(name: str):
     _artifact = {}
 
@@ -344,57 +315,70 @@ def retrieve_available_artifacts():
 
 
 if __name__ == "__main__":
-    github_actions_job_links = get_job_links()
-    available_artifacts = retrieve_available_artifacts()
+    SLACK_REPORT_CHANNEL_ID = os.environ["SLACK_REPORT_CHANNEL"]
 
-    docs = collections.OrderedDict(
-        [
-            ("*.py", "API Examples"),
-            ("*.md", "MD Examples"),
-        ]
+    github_actions_jobs = get_jobs(
+        workflow_run_id=os.environ["GITHUB_RUN_ID"], token=os.environ["ACCESS_REPO_INFO_TOKEN"]
     )
 
-    # This dict will contain all the information relative to each doc test category:
-    # - failed: list of failed tests
-    # - failures: dict in the format 'test': 'error_message'
-    doc_test_results = {
-        v: {
-            "failed": [],
-            "failures": {},
-        }
-        for v in docs.values()
-    }
+    artifact_name_to_job_map = {}
+    for job in github_actions_jobs:
+        for step in job["steps"]:
+            if step["name"].startswith("Test suite reports artifacts: "):
+                artifact_name = step["name"][len("Test suite reports artifacts: ") :]
+                artifact_name_to_job_map[artifact_name] = job
+                break
 
-    # Link to the GitHub Action job
-    doc_test_results["job_link"] = github_actions_job_links.get("run_doctests")
+    available_artifacts = retrieve_available_artifacts()
 
-    artifact_path = available_artifacts["doc_tests_gpu_test_reports"].paths[0]
-    artifact = retrieve_artifact(artifact_path["name"])
-    if "stats" in artifact:
-        failed, success, time_spent = handle_test_results(artifact["stats"])
-        doc_test_results["failures"] = failed
-        doc_test_results["success"] = success
-        doc_test_results["time_spent"] = time_spent[1:-1] + ", "
+    doc_test_results = {}
+    # `artifact_key` is the artifact path
+    for artifact_key, artifact_obj in available_artifacts.items():
+        artifact_path = artifact_obj.paths[0]
+        if not artifact_path["path"].startswith("doc_tests_gpu_test_reports_"):
+            continue
 
-        all_failures = extract_first_line_failure(artifact["failures_short"])
-        for line in artifact["summary_short"].split("\n"):
-            if re.search("FAILED", line):
-                line = line.replace("FAILED ", "")
-                line = line.split()[0].replace("\n", "")
+        # change "_" back to "/" (to show the job name as path)
+        job_name = artifact_path["path"].replace("doc_tests_gpu_test_reports_", "").replace("_", "/")
 
-                if "::" in line:
-                    file_path, test = line.split("::")
-                else:
-                    file_path, test = line, line
+        # This dict (for each job) will contain all the information relative to each doc test job, in particular:
+        #   - failed: list of failed tests
+        #   - failures: dict in the format 'test': 'error_message'
+        job_result = {}
+        doc_test_results[job_name] = job_result
 
-                for file_regex in docs.keys():
-                    if fnmatch(file_path, file_regex):
-                        category = docs[file_regex]
-                        doc_test_results[category]["failed"].append(test)
+        job = artifact_name_to_job_map[artifact_path["path"]]
+        job_result["job_link"] = job["html_url"]
+        job_result["category"] = "Python Examples" if job_name.startswith("src/") else "MD Examples"
 
-                        failure = all_failures[test] if test in all_failures else "N/A"
-                        doc_test_results[category]["failures"][test] = failure
-                        break
+        artifact = retrieve_artifact(artifact_path["path"])
+        if "stats" in artifact:
+            failed, success, time_spent = handle_test_results(artifact["stats"])
+            job_result["n_failures"] = failed
+            job_result["n_success"] = success
+            job_result["time_spent"] = time_spent[1:-1] + ", "
+            job_result["failed"] = []
+            job_result["failures"] = {}
+
+            all_failures = extract_first_line_failure(artifact["failures_short"])
+            for line in artifact["summary_short"].split("\n"):
+                if re.search("FAILED", line):
+                    line = line.replace("FAILED ", "")
+                    line = line.split()[0].replace("\n", "")
+
+                    if "::" in line:
+                        file_path, test = line.split("::")
+                    else:
+                        file_path, test = line, line
+
+                    job_result["failed"].append(test)
+                    failure = all_failures[test] if test in all_failures else "N/A"
+                    job_result["failures"][test] = failure
+
+    # Save and to be uploaded as artifact
+    os.makedirs("doc_test_results", exist_ok=True)
+    with open("doc_test_results/doc_test_results.json", "w", encoding="UTF-8") as fp:
+        json.dump(doc_test_results, fp, ensure_ascii=False, indent=4)
 
     message = Message("🤗 Results of the doc tests.", doc_test_results)
     message.post()

utils/split_doctest_jobs.py

@@ -0,0 +1,96 @@
+# Copyright 2024 The HuggingFace Team. All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+"""
+This script is used to get the files against which we will run doc testing.
+This uses `tests_fetcher.get_all_doctest_files` then groups the test files by their directory paths.
+
+The files in `docs/source/en/model_doc` or `docs/source/en/tasks` are **NOT** grouped together with other files in the
+same directory: the objective is to run doctest against them in independent GitHub Actions jobs.
+
+Assume we are under `transformers` root directory:
+To get a map (dictionary) between directory (or file) paths and the corresponding files
+```bash
+python utils/split_doctest_jobs.py
+```
+or to get a list of lists of directory (or file) paths
+```bash
+python utils/split_doctest_jobs.py --only_return_keys --num_splits 4
+```
+(this is used to allow GitHub Actions to generate more than 256 jobs using matrix)
+"""
+
+import argparse
+from collections import defaultdict
+from pathlib import Path
+
+from tests_fetcher import get_all_doctest_files
+
+
+if __name__ == "__main__":
+    parser = argparse.ArgumentParser()
+    parser.add_argument(
+        "--only_return_keys",
+        action="store_true",
+        help="if to only return the keys (which is a list of list of files' directory or file paths).",
+    )
+    parser.add_argument(
+        "--num_splits",
+        type=int,
+        default=1,
+        help="the number of splits into which the (flat) list of direcotry/file paths will be split. This has effect only if `only_return_keys` is `True`.",
+    )
+    args = parser.parse_args()
+
+    all_doctest_files = get_all_doctest_files()
+
+    raw_test_collection_map = defaultdict(list)
+
+    for file in all_doctest_files:
+        file_dir = "/".join(Path(file).parents[0].parts)
+        raw_test_collection_map[file_dir].append(file)
+
+    refined_test_collection_map = {}
+    for file_dir in raw_test_collection_map.keys():
+        if "docs/source/en/" in file_dir:
+            continue
+        elif file_dir in ["docs/source/en/model_doc", "docs/source/en/tasks"]:
+            for file in raw_test_collection_map[file_dir]:
+                refined_test_collection_map[file] = file
+        else:
+            refined_test_collection_map[file_dir] = " ".join(sorted(raw_test_collection_map[file_dir]))
+
+    sorted_file_dirs = sorted(refined_test_collection_map.keys())
+
+    # # TODO: remove
+    # sorted_file_dirs = sorted_file_dirs[0:4]
+
+    test_collection_map = {}
+    for file_dir in sorted_file_dirs:
+        test_collection_map[file_dir] = refined_test_collection_map[file_dir]
+
+    num_jobs = len(test_collection_map)
+    num_jobs_per_splits = num_jobs // args.num_splits
+
+    file_directory_splits = []
+    end = 0
+    for idx in range(args.num_splits):
+        start = end
+        end = start + num_jobs_per_splits + (1 if idx < num_jobs % args.num_splits else 0)
+        file_directory_splits.append(sorted_file_dirs[start:end])
+
+    if args.only_return_keys:
+        print(file_directory_splits)
+    else:
+        print(dict(test_collection_map))

utils/tests_fetcher.py

@@ -502,7 +502,10 @@ def get_all_doctest_files() -> List[str]:
     """
     py_files = [str(x.relative_to(PATH_TO_REPO)) for x in PATH_TO_REPO.glob("**/*.py")]
     md_files = [str(x.relative_to(PATH_TO_REPO)) for x in PATH_TO_REPO.glob("**/*.md")]
+
     test_files_to_run = py_files + md_files
+    # change to use "/" as path separator
+    test_files_to_run = ["/".join(Path(x).parts) for x in test_files_to_run]
 
     # only include files in `src` or `docs/source/en/`
     test_files_to_run = [x for x in test_files_to_run if x.startswith(("src/", "docs/source/en/"))]