diff --git a/brainscore_language/model_helpers/huggingface.py b/brainscore_language/model_helpers/huggingface.py
index 84576853..89c42f12 100644
--- a/brainscore_language/model_helpers/huggingface.py
+++ b/brainscore_language/model_helpers/huggingface.py
@@ -42,9 +42,8 @@ def __init__(
         self._logger = logging.getLogger(fullname(self))
         self.model_id = model_id
         self.region_layer_mapping = region_layer_mapping
-        self.basemodel = (model if model is not None else AutoModelForCausalLM.from_pretrained(self.model_id))
+        self.basemodel = (model if model is not None else AutoModelForCausalLM.from_pretrained(self.model_id, device_map="auto"))
         self.device = 'cuda' if torch.cuda.is_available() else 'cpu'
-        self.basemodel.to(self.device)
         self.tokenizer = tokenizer if tokenizer is not None else AutoTokenizer.from_pretrained(self.model_id,
                                                                                                truncation_side='left')
         self.current_tokens = None  # keep track of current tokens
@@ -182,6 +181,10 @@ def _tokenize_overflow_aware(self, context, num_previous_context_tokens: int) ->
             context_tokens.pop('num_truncated_tokens')
         if 'overflow_to_sample_mapping' in context_tokens:
             context_tokens.pop('overflow_to_sample_mapping')
+        if 'token_type_ids' in context_tokens:
+            context_tokens.pop('token_type_ids')
+        if self.basemodel.config.is_encoder_decoder:
+            context_tokens['decoder_input_ids'] = context_tokens['input_ids']
         context_tokens.to(self.device)
         return context_tokens, num_new_context_tokens
 
diff --git a/brainscore_language/models/llama/__init__.py b/brainscore_language/models/llama/__init__.py
new file mode 100644
index 00000000..e4b25a50
--- /dev/null
+++ b/brainscore_language/models/llama/__init__.py
@@ -0,0 +1,47 @@
+from brainscore_language import model_registry
+from brainscore_language import ArtificialSubject
+from brainscore_language.model_helpers.huggingface import HuggingfaceSubject
+from transformers import AutoModelForCausalLM, AutoTokenizer, LlamaTokenizer
+
+# layer assignment based on choosing the maximally scoring layer on Pereira2018-encoding
+
+# LLaMA 1 models
+
+model_registry['llama-7b'] = lambda: HuggingfaceSubject(
+    model_id='huggyllama/llama-7b', 
+    region_layer_mapping={ArtificialSubject.RecordingTarget.language_system: 'model.layers.29.post_attention_layernorm'}
+)
+
+model_registry['llama-13b'] = lambda: HuggingfaceSubject(
+    model_id='huggyllama/llama-13b', 
+    region_layer_mapping={ArtificialSubject.RecordingTarget.language_system: 'model.layers.33.post_attention_layernorm'}
+)
+
+model_registry['llama-33b'] = lambda: HuggingfaceSubject(
+    model_id='huggyllama/llama-30b', 
+    region_layer_mapping={ArtificialSubject.RecordingTarget.language_system: 'model.layers.53.post_attention_layernorm'}
+)
+
+# Alpaca models
+
+model_registry['alpaca-7b'] = lambda: HuggingfaceSubject(
+    model_id='chavinlo/alpaca-native', 
+    region_layer_mapping={ArtificialSubject.RecordingTarget.language_system: 'model.layers.29.post_attention_layernorm'}
+)
+
+# Vicuna models
+
+model_registry['vicuna-7b'] = lambda: HuggingfaceSubject(
+    model_id='lmsys/vicuna-7b-v1.3', 
+    region_layer_mapping={ArtificialSubject.RecordingTarget.language_system: 'model.layers.28.post_attention_layernorm'}
+)
+
+model_registry['vicuna-13b'] = lambda: HuggingfaceSubject(
+    model_id='lmsys/vicuna-13b-v1.3', 
+    region_layer_mapping={ArtificialSubject.RecordingTarget.language_system: 'model.layers.33.post_attention_layernorm'}
+)
+
+model_registry['vicuna-33b'] = lambda: HuggingfaceSubject(
+    model_id='lmsys/vicuna-33b-v1.3', 
+    region_layer_mapping={ArtificialSubject.RecordingTarget.language_system: 'model.layers.52.post_attention_layernorm'}
+)
diff --git a/brainscore_language/models/llama/test.py b/brainscore_language/models/llama/test.py
new file mode 100644
index 00000000..5a56787b
--- /dev/null
+++ b/brainscore_language/models/llama/test.py
@@ -0,0 +1,62 @@
+import numpy as np
+import pytest
+
+from brainscore_language import load_model
+from brainscore_language.artificial_subject import ArtificialSubject
+
+
+@pytest.mark.memory_intense
+@pytest.mark.parametrize('model_identifier, expected_reading_times', [
+    ('llama-7b', [9.724511, 12.605466, 3.32503, 0.09871647, 0.725152, 0.04576033, 0.07947908, 0.08976307]),
+    ('llama-13b',[10.53345, 11.900979, 2.576608, 0.09501585, 0.6747948, 0.06707504, 0.07982931, 0.13605802]),
+    ('llama-33b', [11.483265, 12.449862, 1.7104287, 0.10519427, 0.9729844, 0.12699145, 0.23386568, 0.15289368]),
+    ('alpaca-7b', [3.15336514e+01, 1.61361885e+01, 6.20819473e+00, 3.02336123e-02, 4.87159938e-01, 5.48269460e-03, 1.08295875e-02, 1.63752567e-02]),
+    ('vicuna-7b', [1.4193897e+01, 1.4030097e+01, 4.5661983e+00, 1.7538711e-02, 5.8269405e-01, 3.2116382e-03, 8.8979863e-02, 7.2399867e-03]),
+    ('vicuna-13b', [5.1001291e+00, 1.1878480e+01, 7.0294745e-02, 2.8342367e-03, 8.7360293e-03, 6.8028755e-03, 5.5397633e-02, 3.3574910e-03]),
+    ('vicuna-33b', [4.8655987, 14.37647, 1.5682482, 0.02738321, 0.34660488, 0.04076412, 0.0271305, 0.03512227]),
+])
+def test_reading_times(model_identifier, expected_reading_times):
+    model = load_model(model_identifier)
+    text = ['the', 'quick', 'brown', 'fox', 'jumps', 'over', 'the', 'lazy']
+    model.start_behavioral_task(task=ArtificialSubject.Task.reading_times)
+    reading_times = model.digest_text(text)['behavior']
+    np.testing.assert_allclose(reading_times, expected_reading_times, atol=0.01)
+
+
+@pytest.mark.memory_intense
+@pytest.mark.parametrize('model_identifier, expected_next_words', [
+    ('llama-7b', ['j', 'the', 'dog']),
+    ('llama-13b', ['j', 'the', 'dog']),
+    ('llama-33b', ['j', 'the', 'dog']),
+    ('alpaca-7b', ['j', 'the', 'dog']),
+    ('vicuna-7b', ['j', 'the', 'dog']),
+    ('vicuna-13b', ['j', 'the', 'dog']),
+    ('vicuna-33b', ['j', 'the', 'dog']),
+])
+def test_next_word(model_identifier, expected_next_words):
+    model = load_model(model_identifier)
+    text = ['the quick brown fox', 'jumps over', 'the lazy']
+    model.start_behavioral_task(task=ArtificialSubject.Task.next_word)
+    next_word_predictions = model.digest_text(text)['behavior']
+    np.testing.assert_array_equal(next_word_predictions, expected_next_words)
+
+
+@pytest.mark.memory_intense
+@pytest.mark.parametrize('model_identifier, feature_size', [
+    ('llama-7b', 4096),
+    ('llama-13b', 5120),
+    ('llama-33b', 6656),
+    ('alpaca-7b', 4096),
+    ('vicuna-7b', 4096),
+    ('vicuna-13b', 5120),
+    ('vicuna-33b', 6656),
+])
+def test_neural(model_identifier, feature_size):
+    model = load_model(model_identifier)
+    text = ['the quick brown fox', 'jumps over', 'the lazy dog']
+    model.start_neural_recording(recording_target=ArtificialSubject.RecordingTarget.language_system,
+                                 recording_type=ArtificialSubject.RecordingType.fMRI)
+    representations = model.digest_text(text)['neural']
+    assert len(representations['presentation']) == 3
+    np.testing.assert_array_equal(representations['stimulus'], text)
+    assert len(representations['neuroid']) == feature_size
diff --git a/brainscore_language/models/t5/__init__.py b/brainscore_language/models/t5/__init__.py
new file mode 100644
index 00000000..3d8e80c9
--- /dev/null
+++ b/brainscore_language/models/t5/__init__.py
@@ -0,0 +1,116 @@
+from brainscore_language import model_registry
+from brainscore_language import ArtificialSubject
+from brainscore_language.model_helpers.huggingface import HuggingfaceSubject
+from transformers import AutoModelForSeq2SeqLM
+
+# layer assignment based on choosing the maximally scoring layer on Pereira2018-encoding
+
+# T5 models
+
+model_registry['t5-small'] = lambda: HuggingfaceSubject(
+    model_id='google/t5-v1_1-small', 
+    model=AutoModelForSeq2SeqLM.from_pretrained('google/t5-v1_1-small', device_map="auto"),
+    region_layer_mapping={ArtificialSubject.RecordingTarget.language_system: 'decoder.block.6'}
+)
+
+model_registry['t5-base'] = lambda: HuggingfaceSubject(
+    model_id='google/t5-v1_1-base', 
+    model=AutoModelForSeq2SeqLM.from_pretrained('google/t5-v1_1-base', device_map="auto"),
+    region_layer_mapping={ArtificialSubject.RecordingTarget.language_system: 'encoder.block.9'}
+)
+
+model_registry['t5-large'] = lambda: HuggingfaceSubject(
+    model_id='google/t5-v1_1-large', 
+    model=AutoModelForSeq2SeqLM.from_pretrained('google/t5-v1_1-large', device_map="auto"),
+    region_layer_mapping={ArtificialSubject.RecordingTarget.language_system: 'encoder.block.17'}
+)
+
+model_registry['t5-xl'] = lambda: HuggingfaceSubject(
+    model_id='google/t5-v1_1-xl', 
+    model=AutoModelForSeq2SeqLM.from_pretrained('google/t5-v1_1-xl', device_map="auto"),
+    region_layer_mapping={ArtificialSubject.RecordingTarget.language_system: 'decoder.block.2'}
+)
+
+model_registry['t5-xxl'] = lambda: HuggingfaceSubject(
+    model_id='google/t5-v1_1-xxl', 
+    model=AutoModelForSeq2SeqLM.from_pretrained('google/t5-v1_1-xxl', device_map="auto"),
+    region_layer_mapping={ArtificialSubject.RecordingTarget.language_system: 'decoder.block.0'}
+)
+
+# Flan-T5 models
+
+model_registry['flan-t5-small'] = lambda: HuggingfaceSubject(
+    model_id='google/flan-t5-small', 
+    model=AutoModelForSeq2SeqLM.from_pretrained('google/flan-t5-small', device_map="auto"),
+    region_layer_mapping={ArtificialSubject.RecordingTarget.language_system: 'encoder.block.7'}
+)
+
+model_registry['flan-t5-base'] = lambda: HuggingfaceSubject(
+    model_id='google/flan-t5-base', 
+    model=AutoModelForSeq2SeqLM.from_pretrained('google/flan-t5-base', device_map="auto"),
+    region_layer_mapping={ArtificialSubject.RecordingTarget.language_system: 'decoder.block.7'}
+)
+
+model_registry['flan-t5-large'] = lambda: HuggingfaceSubject(
+    model_id='google/flan-t5-large', 
+    model=AutoModelForSeq2SeqLM.from_pretrained('google/flan-t5-large', device_map="auto"),
+    region_layer_mapping={ArtificialSubject.RecordingTarget.language_system: 'encoder.block.18'}
+)
+
+model_registry['flan-t5-xl'] = lambda: HuggingfaceSubject(
+    model_id='google/flan-t5-xl', 
+    model=AutoModelForSeq2SeqLM.from_pretrained('google/flan-t5-xl', device_map="auto"),
+    region_layer_mapping={ArtificialSubject.RecordingTarget.language_system: 'decoder.block.2'}
+)
+
+model_registry['flan-t5-xxl'] = lambda: HuggingfaceSubject(
+    model_id='google/flan-t5-xxl', 
+    model=AutoModelForSeq2SeqLM.from_pretrained('google/flan-t5-xxl', device_map="auto"),
+    region_layer_mapping={ArtificialSubject.RecordingTarget.language_system: 'decoder.block.0'}
+)
+
+# Flan-Alpaca T5 models
+
+model_registry['flan-alpaca-base'] = lambda: HuggingfaceSubject(
+    model_id='declare-lab/flan-alpaca-base', 
+    model=AutoModelForSeq2SeqLM.from_pretrained('declare-lab/flan-alpaca-base', device_map="auto"),
+    region_layer_mapping={ArtificialSubject.RecordingTarget.language_system: 'encoder.block.11'}
+)
+
+model_registry['flan-alpaca-large'] = lambda: HuggingfaceSubject(
+    model_id='declare-lab/flan-alpaca-large', 
+    model=AutoModelForSeq2SeqLM.from_pretrained('declare-lab/flan-alpaca-large', device_map="auto"),
+    region_layer_mapping={ArtificialSubject.RecordingTarget.language_system: 'encoder.block.19'}
+)
+
+model_registry['flan-alpaca-xl'] = lambda: HuggingfaceSubject(
+    model_id='declare-lab/flan-alpaca-xl', 
+    model=AutoModelForSeq2SeqLM.from_pretrained('declare-lab/flan-alpaca-xl', device_map="auto"),
+    region_layer_mapping={ArtificialSubject.RecordingTarget.language_system: 'decoder.block.5'}
+)
+
+model_registry['flan-alpaca-xxl'] = lambda: HuggingfaceSubject(
+    model_id='declare-lab/flan-alpaca-xxl', 
+    model=AutoModelForSeq2SeqLM.from_pretrained('declare-lab/flan-alpaca-xxl', device_map="auto"),
+    region_layer_mapping={ArtificialSubject.RecordingTarget.language_system: 'decoder.block.0'}
+)
+
+# Other models based on T5
+
+model_registry['flan-gpt4all-xl'] = lambda: HuggingfaceSubject(
+    model_id='declare-lab/flan-gpt4all-xl', 
+    model=AutoModelForSeq2SeqLM.from_pretrained('declare-lab/flan-gpt4all-xl', device_map="auto"),
+    region_layer_mapping={ArtificialSubject.RecordingTarget.language_system: 'decoder.block.2'}
+)
+
+model_registry['flan-sharegpt-xl'] = lambda: HuggingfaceSubject(
+    model_id='declare-lab/flan-sharegpt-xl', 
+    model=AutoModelForSeq2SeqLM.from_pretrained('declare-lab/flan-sharegpt-xl', device_map="auto"),
+    region_layer_mapping={ArtificialSubject.RecordingTarget.language_system: 'decoder.block.5'}
+)
+
+model_registry['flan-alpaca-gpt4-xl'] = lambda: HuggingfaceSubject(
+    model_id='declare-lab/flan-alpaca-gpt4-xl', 
+    model=AutoModelForSeq2SeqLM.from_pretrained('declare-lab/flan-alpaca-gpt4-xl', device_map="auto"),
+    region_layer_mapping={ArtificialSubject.RecordingTarget.language_system: 'decoder.block.2'}
+)
diff --git a/brainscore_language/models/t5/requirements.txt b/brainscore_language/models/t5/requirements.txt
new file mode 100644
index 00000000..fa7b252f
--- /dev/null
+++ b/brainscore_language/models/t5/requirements.txt
@@ -0,0 +1,2 @@
+sentencepiece
+protobuf==3.19.4
diff --git a/brainscore_language/models/t5/test.py b/brainscore_language/models/t5/test.py
new file mode 100644
index 00000000..de5a07dc
--- /dev/null
+++ b/brainscore_language/models/t5/test.py
@@ -0,0 +1,92 @@
+import numpy as np
+import pytest
+
+from brainscore_language import load_model
+from brainscore_language.artificial_subject import ArtificialSubject
+
+
+@pytest.mark.memory_intense
+@pytest.mark.parametrize('model_identifier, expected_reading_times', [
+    ('t5-small', [25.646585, 23.780153, 23.018826, 22.344381, 11.96658, 27.054287, 10.594951, 13.187043]),
+    ('t5-base', [7.7039944e-03, 6.8635613e-02, 3.1093130e+01, 1.2913298e+02, 8.5430244e+01, 1.6261120e+01, 8.2980719e+00, 2.9535002e+01]),
+    ('t5-large', [31.604916, 18.852331, 30.816673, 48.99762, 49.006733, 36.088543, 14.189968, 37.781395]),
+    ('t5-xl', [ 5.2831264, 18.823713, 19.249414, 35.212494, 24.10475, 19.929758, 11.064505, 16.397375 ]),
+    ('t5-xxl', [26.934216, 30.064108, 18.61358, 71.8481, 20.456089, 18.108957, 25.52297, 20.845043]),
+    ('flan-t5-small', [4.626572, 5.4074254, 2.9690156, 5.98445, 12.027061, 11.096782, 16.912296, 14.794151]),
+    ('flan-t5-base', [1.8610231, 1.5091983, 2.3265584, 2.5798035, 0.9352376, 2.594869, 3.4819074, 2.7790558]),
+    ('flan-t5-large', [2.2994747, 4.1134634, 1.6111257, 10.103671, 11.365605, 3.37785, 1.4599704, 2.9243639]),
+    ('flan-t5-xl', [2.5323708, 2.9281907, 3.2239344, 10.614168, 7.162341, 3.0385818, 2.9526176, 2.7103176]),
+    ('flan-t5-xxl', [2.3222983, 2.3133714, 2.8529167, 11.162584, 6.798625, 4.742971, 2.9756427, 2.9877827]),
+    ('flan-alpaca-base', [0.5997408, 1.1441187, 1.3299922, 2.1235154, 1.5477583, 0.27742645, 0.3976275, 0.21495701]),
+    ('flan-alpaca-large', [0.03638878, 0.07655565, 0.02087213, 11.400998, 9.982766, 0.82122284, 0.42820516, 0.39627305]),
+    ('flan-alpaca-xl', [3.2593443, 3.6223898, 3.3259575, 12.523176, 6.452489, 5.2135086, 3.7474098, 3.6356025]),
+    ('flan-alpaca-xxl', [2.916435, 5.631528, 3.178902, 11.2796755, 5.902015, 2.294983, 2.8577528, 2.9340065]),
+    ('flan-gpt4all-xl', [6.95467, 8.141007, 6.8901677, 7.149359, 7.247072, 7.390025, 5.7526765, 4.9763246]),
+    ('flan-sharegpt-xl', [3.0441425, 2.9028635, 3.034965, 5.7231064, 2.282282, 2.5237873, 1.0039636, 1.014216]),
+    ('flan-alpaca-gpt4-xl', [5.705884, 6.2532945, 5.6363673, 12.22221, 6.067267, 4.2973313, 4.1460104, 5.088393]),
+])
+def test_reading_times(model_identifier, expected_reading_times):
+    model = load_model(model_identifier)
+    text = ['the', 'quick', 'brown', 'fox', 'jumps', 'over', 'the', 'lazy']
+    model.start_behavioral_task(task=ArtificialSubject.Task.reading_times)
+    reading_times = model.digest_text(text)['behavior']
+    np.testing.assert_allclose(reading_times, expected_reading_times, atol=0.01)
+
+
+@pytest.mark.memory_intense
+@pytest.mark.parametrize('model_identifier, expected_next_words', [
+    ('t5-small', ['in', 'in', 'in']),
+    ('t5-base', ['<extra_id_27>', '</s>', '<extra_id_27>']),
+    ('t5-large', ['<extra_id_11>', '<extra_id_11>', '<extra_id_11>']),
+    ('t5-xl', ['', '', '']),
+    ('t5-xxl', ['', 'ES', ',']),
+    ('flan-t5-small', ['...', '...', '...']),
+    ('flan-t5-base', ['</s>', '...', '</s>']),
+    ('flan-t5-large', ['', '', '']),
+    ('flan-t5-xl', ['', '...', '</s>']),
+    ('flan-t5-xxl', ['</s>', '.', '.']),
+    ('flan-alpaca-base', ['</s>', '</s>', '</s>']),
+    ('flan-alpaca-large', ['', '</s>', '</s>']),
+    ('flan-alpaca-xl', ['', '.', '.']),
+    ('flan-alpaca-xxl', ['.', '.', '.']),
+    ('flan-gpt4all-xl', ['', '', '']),
+    ('flan-sharegpt-xl', ['the', '</s>', '</s>']),
+    ('flan-alpaca-gpt4-xl', ['', '</s>', '</s>']),
+])
+def test_next_word(model_identifier, expected_next_words):
+    model = load_model(model_identifier)
+    text = ['the quick brown fox', 'jumps over', 'the lazy']
+    model.start_behavioral_task(task=ArtificialSubject.Task.next_word)
+    next_word_predictions = model.digest_text(text)['behavior']
+    np.testing.assert_array_equal(next_word_predictions, expected_next_words)
+
+
+@pytest.mark.memory_intense
+@pytest.mark.parametrize('model_identifier, feature_size', [
+    ('t5-small', 512),
+    ('t5-base', 768),
+    ('t5-large', 1024),
+    ('t5-xl', 2048),
+    ('t5-xxl', 4096),
+    ('flan-t5-small', 512),
+    ('flan-t5-base', 768),
+    ('flan-t5-large', 1024),
+    ('flan-t5-xl', 2048),
+    ('flan-t5-xxl', 4096),
+    ('flan-alpaca-base', 768),
+    ('flan-alpaca-large', 1024),
+    ('flan-alpaca-xl', 2048),
+    ('flan-alpaca-xxl', 4096),
+    ('flan-gpt4all-xl', 2048),
+    ('flan-sharegpt-xl', 2048),
+    ('flan-alpaca-gpt4-xl', 2048),
+])
+def test_neural(model_identifier, feature_size):
+    model = load_model(model_identifier)
+    text = ['the quick brown fox', 'jumps over', 'the lazy dog']
+    model.start_neural_recording(recording_target=ArtificialSubject.RecordingTarget.language_system,
+                                 recording_type=ArtificialSubject.RecordingType.fMRI)
+    representations = model.digest_text(text)['neural']
+    assert len(representations['presentation']) == 3
+    np.testing.assert_array_equal(representations['stimulus'], text)
+    assert len(representations['neuroid']) == feature_size
diff --git a/pyproject.toml b/pyproject.toml
index f2237e92..f4cbb4ea 100644
--- a/pyproject.toml
+++ b/pyproject.toml
@@ -22,6 +22,7 @@ dependencies = [
     "transformers>=4.11.3",
     "gensim",
     "joblib",
+    "accelerate",
     # submission dependencies
     "requests"
 ]