Merge branch 'main' into attention-attribution\n\nNeeded to downgrade…

… torch to ~1.12.1 again because of platform issue: pytorch/pytorch#88826

.github/workflows/build.yml

@@ -12,13 +12,13 @@ jobs:
     if: github.actor != 'dependabot[bot]' && github.actor != 'dependabot-preview[bot]'
     strategy:
       matrix:
-        python-version: ["3.8", "3.9", "3.10"]
+        python-version: ["3.8", "3.9", "3.10", "3.11"]
 
     steps:
     - uses: actions/checkout@v3
 
     - name: Set up Python ${{ matrix.python-version }}
-      uses: actions/setup-python@v3
+      uses: actions/setup-python@v4.3.0
       with:
         python-version: ${{ matrix.python-version }}
 
@@ -28,14 +28,14 @@ jobs:
         export PATH="$HOME/.poetry/env:$PATH"
 
     - name: Set up cache
-      uses: actions/cache@v3.0.1
+      uses: actions/cache@v3.0.11
       with:
         path: .venv
         key: venv-${{ matrix.python-version }}-${{ hashFiles('pyproject.toml') }}-${{ hashFiles('poetry.lock') }}
     - name: Install dependencies
       run: |
         poetry config virtualenvs.in-project true
-        poetry install
+        make install-ci
 
     - name: Run style checks
       run: |

.pre-commit-config.yaml

@@ -5,7 +5,7 @@ default_stages: [commit, push]
 
 repos:
   - repo: https://github.com/pre-commit/pre-commit-hooks
-    rev: v4.3.0
+    rev: v4.4.0
     hooks:
       - id: trailing-whitespace
       - id: check-yaml

Makefile

@@ -44,15 +44,16 @@ poetry-remove:
 .PHONY: add-torch-gpu
 add-torch-gpu:
 	poetry run poe upgrade-pip
+	poetry run pip uninstall torch -y
 	poetry run poe torch-cuda11
 
 .PHONY: install
 install:
-	poetry install --no-dev
+	poetry install
 
 .PHONY: install-dev
 install-dev:
-	poetry install --extras all
+	poetry install --all-extras --with lint,docs --sync
 #	-poetry run mypy --install-types --non-interactive ./
 	poetry run pre-commit install
 	poetry run pre-commit autoupdate
@@ -63,6 +64,10 @@ install-gpu: install add-torch-gpu
 .PHONY: install-dev-gpu
 install-dev-gpu: install-dev add-torch-gpu
 
+.PHONY: install-ci
+install-ci:
+	poetry install --with lint
+
 .PHONY: update-deps
 update-deps:
 	poetry lock && poetry export --without-hashes > requirements.txt

inseq/attr/feat/ops/basic_attention.py

@@ -21,7 +21,6 @@
 from captum._utils.typing import TargetType, TensorOrTupleOfTensorsGeneric
 from captum.attr._utils.attribution import Attribution
 from captum.log import log_usage
-from transformers.modeling_outputs import Seq2SeqLMOutput
 
 from ....utils.typing import MultiStepEmbeddingsTensor
 
@@ -91,7 +90,6 @@ def _merge_attention_heads(
         if option == "average":
             return attention.mean(1)
 
-        # TODO: test this, I feel like this method is not doing what we want here
         elif option == "max":
             return attention.max(1)
 
@@ -127,10 +125,6 @@ def _extract_forward_pass_args(
 
         return forward_pass_args
 
-    def _run_forward_pass(self, **forward_args: dict) -> Seq2SeqLMOutput:
-
-        pass
-
 
 class AggregatedAttention(AttentionAttribution):
     """