Merge pull request #108 from joaonadkarni/update-dependencies

Make dependencies more loose

.github/workflows/build_and_test.yml

@@ -8,7 +8,7 @@ jobs:
     runs-on: ubuntu-20.04
     strategy:
       matrix:
-        python-version: [3.6, 3.7, 3.8]
+        python-version: ["3.7", "3.8", "3.9", "3.10", "3.11"]
 
     steps:
     - uses: actions/checkout@v2

.gitignore

@@ -1,3 +1,6 @@
+# ignore files dumped in tests
+tmp/
+
 *.py[cod]
 __pycache__
 

requirements.txt

@@ -1,11 +1,11 @@
-matplotlib~=3.3.1
-numpy~=1.19.1
-ipython~=7.16.1
-scikit-learn~=0.24.2
-seaborn~=0.11.0
-transformers~=4.6.1
-pytest~=6.1.2
-setuptools~=49.6.0
-torch~=1.9.0
-PyYAML==5.4.1
-captum==0.4.1
+matplotlib>=3.3
+numpy>=1.19
+ipython>=7.16
+scikit-learn>=0.24.2,<2
+seaborn>=0.11
+transformers~=4.6
+pytest>=6.1.2
+setuptools>=49.6.0
+torch>=1.9.0,<3
+PyYAML>=6.0
+captum~=0.4.1

setup.py

@@ -52,6 +52,8 @@ def read(*names, **kwargs):
         'Programming Language :: Python :: 3.7',
         'Programming Language :: Python :: 3.8',
         'Programming Language :: Python :: 3.9',
+        'Programming Language :: Python :: 3.10',
+        'Programming Language :: Python :: 3.11',
         'Programming Language :: Python :: Implementation :: CPython',
         'Programming Language :: Python :: Implementation :: PyPy',
         'Topic :: Utilities',
@@ -63,12 +65,12 @@ def read(*names, **kwargs):
     keywords=[
         'Natural Language Processing', 'Explainable AI', 'keyword3',
     ],
-    python_requires='!=3.0.*, !=3.1.*, !=3.2.*, !=3.3.*, !=3.4.*',
+    python_requires='!=3.0.*, !=3.1.*, !=3.2.*, !=3.3.*, !=3.4.*, !=3.5.*, !=3.6.*',
     install_requires=[
         "transformers ~= 4.2",
-        "seaborn ~= 0.11",
-        "scikit-learn~=0.23",
-        "PyYAML~=5.4",
+        "seaborn >= 0.11",
+        "scikit-learn>=0.23,<2",
+        "PyYAML>=6.0",
         "captum ~= 0.4"
     ],
     extras_require={

src/ecco/lm.py

@@ -199,12 +199,12 @@ def generate(self, input_str: str,
             raise ValueError(
                 "max_length set to {} while input token has more tokens ({}). Consider increasing max_length" \
                     .format(max_length, cur_len))
-        
+
         # Get decoder input ids
         if self.model_type == 'enc-dec': # FIXME: only done because causal LMs like GPT-2 have the _prepare_decoder_input_ids_for_generation method but do not use it
             assert len(input_ids.size()) == 2 # will break otherwise
             if version.parse(transformers.__version__) >= version.parse('4.13'):
-                
+
                 # following the code in https://github.com/huggingface/transformers/blob/d0c1aebea467af499331234e7b285a6bf91ea073/tests/generation/test_utils.py#L2099
                 model_kwargs = self.model._prepare_encoder_decoder_kwargs_for_generation(input_ids, {})
                 decoder_input_ids, model_kwargs = self.model._prepare_decoder_input_ids_for_generation(
@@ -649,7 +649,7 @@ def display_input_sequence(self, input_ids):
                     parentDiv: '{viz_id}',
                     data: {json.dumps(data)},
                     tokenization_config: {json.dumps(self.model_config['tokenizer_config'])}
-            
+
             }})
          }}, function (err) {{
             console.log(err);
@@ -729,10 +729,11 @@ def sample_output_token(scores, do_sample, temperature, top_k, top_p):
         if temperature != 1.0:
             scores = scores / temperature
         # Top-p/top-k filtering
-        next_token_logscores = transformers.generation_utils. \
-            top_k_top_p_filtering(scores,
-                                  top_k=top_k,
-                                  top_p=top_p)
+        if version.parse(transformers.__version__) >= version.parse('4.25.1'):
+            top_k_top_p_filtering_fn = transformers.generation.utils.top_k_top_p_filtering
+        else:
+            top_k_top_p_filtering_fn = transformers.generation_utils.top_k_top_p_filtering
+        next_token_logscores = top_k_top_p_filtering_fn(scores, top_k=top_k, top_p=top_p)
         # Sample
         probs = F.softmax(next_token_logscores, dim=-1)