Unverified Commit 205bc415 authored by Jason Phang's avatar Jason Phang Committed by GitHub
Browse files

Fix GPT-NeoX-20B past handling, attention computation (#17811)

* Fix GPT-NeoX-20B past handling, swap attention computation to hopefully avoid NaN, update docs

* 20B tests
parent 692e61e9
...@@ -38,32 +38,28 @@ class GPTNeoXConfig(PretrainedConfig): ...@@ -38,32 +38,28 @@ class GPTNeoXConfig(PretrainedConfig):
Args: Args:
vocab_size (`int`, *optional*, defaults to 30522): vocab_size (`int`, *optional*, defaults to 50432):
Vocabulary size of the GPTNeoX model. Defines the number of different tokens that can be represented by the Vocabulary size of the GPTNeoX model. Defines the number of different tokens that can be represented by the
`inputs_ids` passed when calling [`GPTNeoXModel`]. `inputs_ids` passed when calling [`GPTNeoXModel`].
hidden_size (`int`, *optional*, defaults to 768): hidden_size (`int`, *optional*, defaults to 6144):
Dimension of the encoder layers and the pooler layer. Dimension of the encoder layers and the pooler layer.
num_hidden_layers (`int`, *optional*, defaults to 12): num_hidden_layers (`int`, *optional*, defaults to 44):
Number of hidden layers in the Transformer encoder. Number of hidden layers in the Transformer encoder.
num_attention_heads (`int`, *optional*, defaults to 12): num_attention_heads (`int`, *optional*, defaults to 64):
Number of attention heads for each attention layer in the Transformer encoder. Number of attention heads for each attention layer in the Transformer encoder.
intermediate_size (`int`, *optional*, defaults to 3072): intermediate_size (`int`, *optional*, defaults to 24576):
Dimension of the "intermediate" (i.e., feed-forward) layer in the Transformer encoder. Dimension of the "intermediate" (i.e., feed-forward) layer in the Transformer encoder.
hidden_act (`str` or `function`, *optional*, defaults to `"gelu"`): hidden_act (`str` or `function`, *optional*, defaults to `"gelu"`):
The non-linear activation function (function or string) in the encoder and pooler. If string, `"gelu"`, The non-linear activation function (function or string) in the encoder and pooler. If string, `"gelu"`,
`"relu"`, `"selu"` and `"gelu_new"` are supported. `"relu"`, `"selu"` and `"gelu_new"` are supported.
hidden_dropout_prob (`float`, *optional*, defaults to 0.1):
The dropout probabilitiy for all fully connected layers in the embeddings, encoder, and pooler.
attention_probs_dropout_prob (`float`, *optional*, defaults to 0.1):
The dropout ratio for the attention probabilities.
rotary_pct (`float`, *optional*, defaults to 0.25): rotary_pct (`float`, *optional*, defaults to 0.25):
percentage of hidden dimensions to allocate to rotary embeddings percentage of hidden dimensions to allocate to rotary embeddings
rotary_emb_base (`int`, *optional*, defaults to 10000) rotary_emb_base (`int`, *optional*, defaults to 10000)
base for computing rotary embeddings frequency base for computing rotary embeddings frequency
max_position_embeddings (`int`, *optional*, defaults to 512): max_position_embeddings (`int`, *optional*, defaults to 2048):
The maximum sequence length that this model might ever be used with. Typically set this to something large The maximum sequence length that this model might ever be used with. Typically set this to something large
just in case (e.g., 512 or 1024 or 2048). just in case (e.g., 512 or 1024 or 2048).
initializer_range (`float`, *optional*, defaults to 0.02): initializer_range (`float`, *optional*, defaults to 1e-5):
The standard deviation of the truncated_normal_initializer for initializing all weight matrices. The standard deviation of the truncated_normal_initializer for initializing all weight matrices.
layer_norm_eps (`float`, *optional*, defaults to 1e-12): layer_norm_eps (`float`, *optional*, defaults to 1e-12):
The epsilon used by the layer normalization layers. The epsilon used by the layer normalization layers.
...@@ -94,8 +90,6 @@ class GPTNeoXConfig(PretrainedConfig): ...@@ -94,8 +90,6 @@ class GPTNeoXConfig(PretrainedConfig):
num_attention_heads=64, num_attention_heads=64,
intermediate_size=24576, intermediate_size=24576,
hidden_act="gelu", hidden_act="gelu",
hidden_dropout_prob=0.1,
attention_probs_dropout_prob=0.1,
rotary_pct=0.25, rotary_pct=0.25,
rotary_emb_base=10000, rotary_emb_base=10000,
max_position_embeddings=2048, max_position_embeddings=2048,
...@@ -115,8 +109,6 @@ class GPTNeoXConfig(PretrainedConfig): ...@@ -115,8 +109,6 @@ class GPTNeoXConfig(PretrainedConfig):
self.num_attention_heads = num_attention_heads self.num_attention_heads = num_attention_heads
self.intermediate_size = intermediate_size self.intermediate_size = intermediate_size
self.hidden_act = hidden_act self.hidden_act = hidden_act
self.hidden_dropout_prob = hidden_dropout_prob
self.attention_probs_dropout_prob = attention_probs_dropout_prob
self.rotary_pct = rotary_pct self.rotary_pct = rotary_pct
self.rotary_emb_base = rotary_emb_base self.rotary_emb_base = rotary_emb_base
self.initializer_range = initializer_range self.initializer_range = initializer_range
......
...@@ -195,7 +195,20 @@ class GPTNeoXAttention(nn.Module): ...@@ -195,7 +195,20 @@ class GPTNeoXAttention(nn.Module):
query = query.view(batch_size * num_attention_heads, query_length, attn_head_size) query = query.view(batch_size * num_attention_heads, query_length, attn_head_size)
key = key.view(batch_size * num_attention_heads, key_length, attn_head_size) key = key.view(batch_size * num_attention_heads, key_length, attn_head_size)
attn_scores = torch.einsum("bik,bjk->bij", query, key) / self.norm_factor attn_scores = torch.zeros(
batch_size * num_attention_heads,
query_length,
key_length,
dtype=query.dtype,
device=key.device,
)
attn_scores = torch.baddbmm(
attn_scores,
query,
key.transpose(1, 2),
beta=1.0,
alpha=(1.0 / self.norm_factor),
)
attn_scores = attn_scores.view(batch_size, num_attention_heads, query_length, key_length) attn_scores = attn_scores.view(batch_size, num_attention_heads, query_length, key_length)
mask_value = torch.finfo(attn_scores.dtype).min mask_value = torch.finfo(attn_scores.dtype).min
...@@ -637,7 +650,7 @@ class GPTNeoXForCausalLM(GPTNeoXPreTrainedModel): ...@@ -637,7 +650,7 @@ class GPTNeoXForCausalLM(GPTNeoXPreTrainedModel):
attention_mask = input_ids.new_ones(input_shape) attention_mask = input_ids.new_ones(input_shape)
# cut decoder_input_ids if past is used # cut decoder_input_ids if past is used
if past is not None: if past and past[0] is not None:
input_ids = input_ids[:, -1:] input_ids = input_ids[:, -1:]
return {"input_ids": input_ids, "attention_mask": attention_mask, "past_key_values": past} return {"input_ids": input_ids, "attention_mask": attention_mask, "past_key_values": past}
......
...@@ -226,6 +226,10 @@ class GPTNeoXModelTest(ModelTesterMixin, unittest.TestCase): ...@@ -226,6 +226,10 @@ class GPTNeoXModelTest(ModelTesterMixin, unittest.TestCase):
config_and_inputs = self.model_tester.prepare_config_and_inputs() config_and_inputs = self.model_tester.prepare_config_and_inputs()
self.model_tester.create_and_check_for_causal_lm(*config_and_inputs) self.model_tester.create_and_check_for_causal_lm(*config_and_inputs)
@unittest.skip(reason="Feed forward chunking is not implemented")
def test_feed_forward_chunking(self):
pass
@slow @slow
def test_model_from_pretrained(self): def test_model_from_pretrained(self):
for model_name in GPT_NEOX_PRETRAINED_MODEL_ARCHIVE_LIST[:1]: for model_name in GPT_NEOX_PRETRAINED_MODEL_ARCHIVE_LIST[:1]:
...@@ -247,7 +251,7 @@ class GPTNeoXModelIntegrationTest(unittest.TestCase): ...@@ -247,7 +251,7 @@ class GPTNeoXModelIntegrationTest(unittest.TestCase):
self.assertEqual(output.shape, expected_shape) self.assertEqual(output.shape, expected_shape)
expected_slice = torch.tensor( expected_slice = torch.tensor(
[[[33.8045, 2.3958, 34.2816], [63.7805, 4.8332, 63.5882], [66.9116, 5.2198, 63.1185]]] [[[33.5938, 2.3789, 34.0312], [63.4688, 4.8164, 63.3438], [66.8750, 5.2422, 63.0625]]]
) )
self.assertTrue(torch.allclose(output[:, :3, :3], expected_slice, atol=1e-4)) self.assertTrue(torch.allclose(output[:, :3, :3], expected_slice, atol=1e-4))
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment