From dff609a8e4aaf6181f47ef03eae109b901755a25 Mon Sep 17 00:00:00 2001 From: AutoJenkins Date: Wed, 29 Nov 2023 06:10:23 -0500 Subject: [PATCH] add gpt to models --- brainscore_language/models/gpt/__init__.py | 3 +++ brainscore_language/models/gpt/test.py | 11 ++++++++--- 2 files changed, 11 insertions(+), 3 deletions(-) diff --git a/brainscore_language/models/gpt/__init__.py b/brainscore_language/models/gpt/__init__.py index 716f255d..3f3f20e2 100644 --- a/brainscore_language/models/gpt/__init__.py +++ b/brainscore_language/models/gpt/__init__.py @@ -31,3 +31,6 @@ model_registry['gpt-neo-1.3B'] = lambda: HuggingfaceSubject(model_id='EleutherAI/gpt-neo-1.3B', region_layer_mapping={ ArtificialSubject.RecordingTarget.language_system: 'transformer.h.18'}) + +model_registry['gpt-j-6b'] = lambda: HuggingfaceSubject(model_id='EleutherAI/gpt-j-6b', region_layer_mapping={ + ArtificialSubject.RecordingTarget.language_system: 'transformer.h.26'}) diff --git a/brainscore_language/models/gpt/test.py b/brainscore_language/models/gpt/test.py index 7cdcf0bd..4b57df3a 100644 --- a/brainscore_language/models/gpt/test.py +++ b/brainscore_language/models/gpt/test.py @@ -24,7 +24,10 @@ ('gpt-neo-2.7B', [np.nan, 15.07522869, 3.6358602 , 0.04999408, 1.42219079, 0.0399301 , 0.02614061, 0.02547451]), ('gpt-neo-1.3B', [np.nan, 15.36009979, 5.54412651, 0.11744193, 0.60116327, - 0.04266951, 0.08952015, 0.09213546]) + 0.04266951, 0.08952015, 0.09213546]), + ('gpt-j-6b', [np.nan, 1.335768e+01, 1.618860e+00, 1.210055e-02, + 3.879260e-01, 4.501237e-02, 5.057606e-02, 9.547059e-02]), + ]) def test_reading_times(model_identifier, expected_reading_times): model = load_model(model_identifier) @@ -44,7 +47,8 @@ def test_reading_times(model_identifier, expected_reading_times): ('gpt2-xl', ['jumps', 'the', 'dog']), ('gpt-neo-125m', [',', 'the', 'dog']), ('gpt-neo-2.7B', ['jumps', 'the', 'dog']), - ('gpt-neo-1.3B', ['jumps', 'the', 'dog']) + ('gpt-neo-1.3B', ['jumps', 'the', 'dog']), + ('gpt-j-6b', ['jumps', 'the', 'dog']), ]) def test_next_word(model_identifier, expected_next_words): model = load_model(model_identifier) @@ -64,7 +68,8 @@ def test_next_word(model_identifier, expected_next_words): ('gpt2-xl', 1600), ('gpt-neo-125m', 768), ('gpt-neo-1.3B', 2048), - ('gpt-neo-2.7B', 2560) + ('gpt-neo-2.7B', 2560), + ('gpt-j-6b', 4096), ]) def test_neural(model_identifier, feature_size): model = load_model(model_identifier)