🎨 format for readability
Browse filesSigned-off-by: peter szemraj <peterszemraj@gmail.com>
- adapt_tokenizer.py +8 -5
- attention.py +287 -70
- blocks.py +58 -11
- configuration_mpt.py +103 -28
- hf_prefixlm_converter.py +440 -102
- meta_init_context.py +26 -10
- norm.py +67 -17
- param_init_fns.py +288 -52
adapt_tokenizer.py
CHANGED
@@ -1,8 +1,10 @@
|
|
1 |
from typing import Union
|
2 |
from transformers import AutoTokenizer, PreTrainedTokenizer, PreTrainedTokenizerFast
|
|
|
3 |
Tokenizer = Union[PreTrainedTokenizer, PreTrainedTokenizerFast]
|
4 |
NUM_SENTINEL_TOKENS: int = 100
|
5 |
|
|
|
6 |
def adapt_tokenizer_for_denoising(tokenizer: Tokenizer):
|
7 |
"""Adds sentinel tokens and padding token (if missing).
|
8 |
|
@@ -12,16 +14,17 @@ def adapt_tokenizer_for_denoising(tokenizer: Tokenizer):
|
|
12 |
All added tokens are added as special tokens. No tokens are
|
13 |
added if sentinel tokens and padding token already exist.
|
14 |
"""
|
15 |
-
sentinels_to_add = [f
|
16 |
tokenizer.add_tokens(sentinels_to_add, special_tokens=True)
|
17 |
if tokenizer.pad_token is None:
|
18 |
-
tokenizer.add_tokens(
|
19 |
-
tokenizer.pad_token =
|
20 |
assert tokenizer.pad_token_id is not None
|
21 |
-
sentinels =
|
22 |
_sentinel_token_ids = tokenizer(sentinels, add_special_tokens=False).input_ids
|
23 |
tokenizer.sentinel_token_ids = _sentinel_token_ids
|
24 |
|
|
|
25 |
class AutoTokenizerForMOD(AutoTokenizer):
|
26 |
"""AutoTokenizer + Adaptation for MOD.
|
27 |
|
@@ -38,4 +41,4 @@ class AutoTokenizerForMOD(AutoTokenizer):
|
|
38 |
"""See `AutoTokenizer.from_pretrained` docstring."""
|
39 |
tokenizer = super().from_pretrained(*args, **kwargs)
|
40 |
adapt_tokenizer_for_denoising(tokenizer)
|
41 |
-
return tokenizer
|
|
|
1 |
from typing import Union
|
2 |
from transformers import AutoTokenizer, PreTrainedTokenizer, PreTrainedTokenizerFast
|
3 |
+
|
4 |
Tokenizer = Union[PreTrainedTokenizer, PreTrainedTokenizerFast]
|
5 |
NUM_SENTINEL_TOKENS: int = 100
|
6 |
|
7 |
+
|
8 |
def adapt_tokenizer_for_denoising(tokenizer: Tokenizer):
|
9 |
"""Adds sentinel tokens and padding token (if missing).
|
10 |
|
|
|
14 |
All added tokens are added as special tokens. No tokens are
|
15 |
added if sentinel tokens and padding token already exist.
|
16 |
"""
|
17 |
+
sentinels_to_add = [f"<extra_id_{i}>" for i in range(NUM_SENTINEL_TOKENS)]
|
18 |
tokenizer.add_tokens(sentinels_to_add, special_tokens=True)
|
19 |
if tokenizer.pad_token is None:
|
20 |
+
tokenizer.add_tokens("<pad>", special_tokens=True)
|
21 |
+
tokenizer.pad_token = "<pad>"
|
22 |
assert tokenizer.pad_token_id is not None
|
23 |
+
sentinels = "".join([f"<extra_id_{i}>" for i in range(NUM_SENTINEL_TOKENS)])
|
24 |
_sentinel_token_ids = tokenizer(sentinels, add_special_tokens=False).input_ids
|
25 |
tokenizer.sentinel_token_ids = _sentinel_token_ids
|
26 |
|
27 |
+
|
28 |
class AutoTokenizerForMOD(AutoTokenizer):
|
29 |
"""AutoTokenizer + Adaptation for MOD.
|
30 |
|
|
|
41 |
"""See `AutoTokenizer.from_pretrained` docstring."""
|
42 |
tokenizer = super().from_pretrained(*args, **kwargs)
|
43 |
adapt_tokenizer_for_denoising(tokenizer)
|
44 |
+
return tokenizer
|
attention.py
CHANGED
@@ -8,18 +8,37 @@ from einops import rearrange
|
|
8 |
from torch import nn
|
9 |
from .norm import LPLayerNorm
|
10 |
|
11 |
-
|
|
|
|
|
|
|
12 |
if original_is_causal and num_query_tokens != num_key_tokens:
|
13 |
if num_query_tokens != 1:
|
14 |
-
raise NotImplementedError(
|
|
|
|
|
15 |
else:
|
16 |
return False
|
17 |
return original_is_causal
|
18 |
|
19 |
-
|
20 |
-
|
21 |
-
|
22 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
23 |
min_val = torch.finfo(q.dtype).min
|
24 |
(b, _, s_q, d) = q.shape
|
25 |
s_k = k.size(-1)
|
@@ -27,13 +46,27 @@ def scaled_multihead_dot_product_attention(query, key, value, n_heads, softmax_s
|
|
27 |
softmax_scale = 1 / math.sqrt(d)
|
28 |
attn_weight = q.matmul(k) * softmax_scale
|
29 |
if attn_bias is not None:
|
30 |
-
if
|
31 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
32 |
attn_weight = attn_weight + attn_bias
|
33 |
if key_padding_mask is not None:
|
34 |
if attn_bias is not None:
|
35 |
-
warnings.warn(
|
36 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
37 |
if is_causal:
|
38 |
s = max(s_q, s_k)
|
39 |
causal_mask = attn_weight.new_ones(s, s, dtype=torch.float16)
|
@@ -44,74 +77,146 @@ def scaled_multihead_dot_product_attention(query, key, value, n_heads, softmax_s
|
|
44 |
attn_weight = attn_weight.masked_fill(causal_mask.view(1, 1, s_q, s_k), min_val)
|
45 |
attn_weight = torch.softmax(attn_weight, dim=-1)
|
46 |
if dropout_p:
|
47 |
-
attn_weight = torch.nn.functional.dropout(
|
|
|
|
|
48 |
out = attn_weight.matmul(v)
|
49 |
-
out = rearrange(out,
|
50 |
if needs_weights:
|
51 |
return (out, attn_weight)
|
52 |
return (out, None)
|
53 |
|
|
|
54 |
def check_valid_inputs(*tensors, valid_dtypes=[torch.float16, torch.bfloat16]):
|
55 |
for tensor in tensors:
|
56 |
if tensor.dtype not in valid_dtypes:
|
57 |
-
raise TypeError(
|
|
|
|
|
58 |
if not tensor.is_cuda:
|
59 |
-
raise TypeError(
|
|
|
|
|
|
|
60 |
|
61 |
-
def flash_attn_fn(
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
62 |
try:
|
63 |
from flash_attn import bert_padding, flash_attn_interface
|
64 |
except:
|
65 |
-
raise RuntimeError(
|
66 |
check_valid_inputs(query, key, value)
|
67 |
if attn_bias is not None:
|
68 |
-
raise NotImplementedError(f
|
69 |
(batch_size, seqlen) = query.shape[:2]
|
70 |
if key_padding_mask is None:
|
71 |
key_padding_mask = torch.ones_like(key[:, :, 0], dtype=torch.bool)
|
72 |
-
query_padding_mask = key_padding_mask[:, -query.size(1):]
|
73 |
-
(query_unpad, indices_q, cu_seqlens_q, max_seqlen_q) = bert_padding.unpad_input(
|
74 |
-
|
75 |
-
|
76 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
77 |
(value_unpad, _, _, _) = bert_padding.unpad_input(value, key_padding_mask)
|
78 |
-
value_unpad = rearrange(
|
|
|
|
|
79 |
if multiquery:
|
80 |
key_unpad = key_unpad.expand(key_unpad.size(0), n_heads, key_unpad.size(-1))
|
81 |
-
value_unpad = value_unpad.expand(
|
|
|
|
|
82 |
dropout_p = dropout_p if training else 0.0
|
83 |
reset_is_causal = _reset_is_causal(query.size(1), key.size(1), is_causal)
|
84 |
-
output_unpad = flash_attn_interface.flash_attn_unpadded_func(
|
85 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
86 |
return (output, None)
|
87 |
|
88 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
89 |
try:
|
90 |
from flash_attn import flash_attn_triton
|
91 |
except:
|
92 |
-
raise RuntimeError(
|
|
|
|
|
93 |
check_valid_inputs(query, key, value)
|
94 |
if dropout_p:
|
95 |
-
raise NotImplementedError(f
|
96 |
if needs_weights:
|
97 |
-
raise NotImplementedError(f
|
98 |
if key_padding_mask is not None:
|
99 |
-
warnings.warn(
|
|
|
|
|
|
|
|
|
|
|
|
|
100 |
(b_size, s_k) = key_padding_mask.shape[:2]
|
101 |
if attn_bias is None:
|
102 |
attn_bias = query.new_zeros(b_size, 1, 1, s_k)
|
103 |
-
attn_bias = attn_bias.masked_fill(
|
104 |
-
|
105 |
-
|
106 |
-
|
|
|
|
|
107 |
if multiquery:
|
108 |
key = key.expand(*key.shape[:2], n_heads, key.size(-1))
|
109 |
value = value.expand(*value.shape[:2], n_heads, value.size(-1))
|
110 |
reset_is_causal = _reset_is_causal(query.size(1), key.size(1), is_causal)
|
111 |
-
attn_output = flash_attn_triton.flash_attn_func(
|
|
|
|
|
112 |
output = attn_output.view(*attn_output.shape[:2], -1)
|
113 |
return (output, None)
|
114 |
|
|
|
115 |
class MultiheadAttention(nn.Module):
|
116 |
"""Multi-head self attention.
|
117 |
|
@@ -119,7 +224,18 @@ class MultiheadAttention(nn.Module):
|
|
119 |
additive bias.
|
120 |
"""
|
121 |
|
122 |
-
def __init__(
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
123 |
super().__init__()
|
124 |
self.attn_impl = attn_impl
|
125 |
self.clip_qkv = clip_qkv
|
@@ -137,21 +253,38 @@ class MultiheadAttention(nn.Module):
|
|
137 |
layernorm_class = LPLayerNorm if low_precision_layernorm else nn.LayerNorm
|
138 |
self.q_ln = layernorm_class(self.d_model, device=device)
|
139 |
self.k_ln = layernorm_class(self.d_model, device=device)
|
140 |
-
if self.attn_impl ==
|
141 |
self.attn_fn = flash_attn_fn
|
142 |
-
elif self.attn_impl ==
|
143 |
self.attn_fn = triton_flash_attn_fn
|
144 |
-
warnings.warn(
|
145 |
-
|
|
|
|
|
|
|
|
|
|
|
146 |
self.attn_fn = scaled_multihead_dot_product_attention
|
147 |
if torch.cuda.is_available():
|
148 |
-
warnings.warn(
|
|
|
|
|
|
|
|
|
149 |
else:
|
150 |
-
raise ValueError(f
|
151 |
self.out_proj = nn.Linear(self.d_model, self.d_model, device=device)
|
152 |
self.out_proj._is_residual = True
|
153 |
|
154 |
-
def forward(
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
155 |
qkv = self.Wqkv(x)
|
156 |
if self.clip_qkv:
|
157 |
qkv.clamp_(min=-self.clip_qkv, max=self.clip_qkv)
|
@@ -167,10 +300,23 @@ class MultiheadAttention(nn.Module):
|
|
167 |
value = torch.cat([past_key_value[1], value], dim=1)
|
168 |
past_key_value = (key, value)
|
169 |
if attn_bias is not None:
|
170 |
-
attn_bias = attn_bias[:, :, -query.size(1):, -key.size(1):]
|
171 |
-
(context, attn_weights) = self.attn_fn(
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
172 |
return (self.out_proj(context), attn_weights, past_key_value)
|
173 |
|
|
|
174 |
class MultiQueryAttention(nn.Module):
|
175 |
"""Multi-Query self attention.
|
176 |
|
@@ -178,7 +324,18 @@ class MultiQueryAttention(nn.Module):
|
|
178 |
additive bias.
|
179 |
"""
|
180 |
|
181 |
-
def __init__(
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
182 |
super().__init__()
|
183 |
self.attn_impl = attn_impl
|
184 |
self.clip_qkv = clip_qkv
|
@@ -197,25 +354,44 @@ class MultiQueryAttention(nn.Module):
|
|
197 |
layernorm_class = LPLayerNorm if low_precision_layernorm else nn.LayerNorm
|
198 |
self.q_ln = layernorm_class(d_model, device=device)
|
199 |
self.k_ln = layernorm_class(self.head_dim, device=device)
|
200 |
-
if self.attn_impl ==
|
201 |
self.attn_fn = flash_attn_fn
|
202 |
-
elif self.attn_impl ==
|
203 |
self.attn_fn = triton_flash_attn_fn
|
204 |
-
warnings.warn(
|
205 |
-
|
|
|
|
|
|
|
|
|
|
|
206 |
self.attn_fn = scaled_multihead_dot_product_attention
|
207 |
if torch.cuda.is_available():
|
208 |
-
warnings.warn(
|
|
|
|
|
|
|
|
|
209 |
else:
|
210 |
-
raise ValueError(f
|
211 |
self.out_proj = nn.Linear(self.d_model, self.d_model, device=device)
|
212 |
self.out_proj._is_residual = True
|
213 |
|
214 |
-
def forward(
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
215 |
qkv = self.Wqkv(x)
|
216 |
if self.clip_qkv:
|
217 |
qkv.clamp_(min=-self.clip_qkv, max=self.clip_qkv)
|
218 |
-
(query, key, value) = qkv.split(
|
|
|
|
|
219 |
key_padding_mask = attention_mask
|
220 |
if self.qk_ln:
|
221 |
dtype = query.dtype
|
@@ -227,14 +403,30 @@ class MultiQueryAttention(nn.Module):
|
|
227 |
value = torch.cat([past_key_value[1], value], dim=1)
|
228 |
past_key_value = (key, value)
|
229 |
if attn_bias is not None:
|
230 |
-
attn_bias = attn_bias[:, :, -query.size(1):, -key.size(1):]
|
231 |
-
(context, attn_weights) = self.attn_fn(
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
232 |
return (self.out_proj(context), attn_weights, past_key_value)
|
233 |
|
234 |
-
|
235 |
-
|
|
|
|
|
|
|
236 |
return None
|
237 |
-
elif attn_impl in [
|
238 |
if alibi:
|
239 |
if (prefix_lm or not causal) or use_sequence_id:
|
240 |
return (1, n_heads, seq_len, seq_len)
|
@@ -243,18 +435,31 @@ def attn_bias_shape(attn_impl, n_heads, seq_len, alibi, prefix_lm, causal, use_s
|
|
243 |
return (1, 1, seq_len, seq_len)
|
244 |
return None
|
245 |
else:
|
246 |
-
raise ValueError(f
|
|
|
247 |
|
248 |
-
def build_attn_bias(
|
249 |
-
|
|
|
|
|
250 |
return None
|
251 |
-
elif attn_impl in [
|
252 |
if alibi:
|
253 |
(device, dtype) = (attn_bias.device, attn_bias.dtype)
|
254 |
-
attn_bias = attn_bias.add(
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
255 |
return attn_bias
|
256 |
else:
|
257 |
-
raise ValueError(f
|
|
|
258 |
|
259 |
def gen_slopes(n_heads, alibi_bias_max=8, device=None):
|
260 |
_n_heads = 2 ** math.ceil(math.log2(n_heads))
|
@@ -265,12 +470,24 @@ def gen_slopes(n_heads, alibi_bias_max=8, device=None):
|
|
265 |
slopes = torch.concat([slopes[1::2], slopes[::2]])[:n_heads]
|
266 |
return slopes.view(1, n_heads, 1, 1)
|
267 |
|
268 |
-
|
269 |
-
|
|
|
|
|
|
|
|
|
|
|
270 |
if full:
|
271 |
-
alibi_bias = alibi_bias - torch.arange(
|
|
|
|
|
272 |
alibi_bias = alibi_bias.abs().mul(-1)
|
273 |
slopes = gen_slopes(n_heads, alibi_bias_max, device=device)
|
274 |
alibi_bias = alibi_bias * slopes
|
275 |
return alibi_bias.to(dtype=dtype)
|
276 |
-
|
|
|
|
|
|
|
|
|
|
|
|
8 |
from torch import nn
|
9 |
from .norm import LPLayerNorm
|
10 |
|
11 |
+
|
12 |
+
def _reset_is_causal(
|
13 |
+
num_query_tokens: int, num_key_tokens: int, original_is_causal: bool
|
14 |
+
):
|
15 |
if original_is_causal and num_query_tokens != num_key_tokens:
|
16 |
if num_query_tokens != 1:
|
17 |
+
raise NotImplementedError(
|
18 |
+
"MPT does not support query and key with different number of tokens, unless number of query tokens is 1."
|
19 |
+
)
|
20 |
else:
|
21 |
return False
|
22 |
return original_is_causal
|
23 |
|
24 |
+
|
25 |
+
def scaled_multihead_dot_product_attention(
|
26 |
+
query,
|
27 |
+
key,
|
28 |
+
value,
|
29 |
+
n_heads,
|
30 |
+
softmax_scale=None,
|
31 |
+
attn_bias=None,
|
32 |
+
key_padding_mask=None,
|
33 |
+
is_causal=False,
|
34 |
+
dropout_p=0.0,
|
35 |
+
training=False,
|
36 |
+
needs_weights=False,
|
37 |
+
multiquery=False,
|
38 |
+
):
|
39 |
+
q = rearrange(query, "b s (h d) -> b h s d", h=n_heads)
|
40 |
+
k = rearrange(key, "b s (h d) -> b h d s", h=1 if multiquery else n_heads)
|
41 |
+
v = rearrange(value, "b s (h d) -> b h s d", h=1 if multiquery else n_heads)
|
42 |
min_val = torch.finfo(q.dtype).min
|
43 |
(b, _, s_q, d) = q.shape
|
44 |
s_k = k.size(-1)
|
|
|
46 |
softmax_scale = 1 / math.sqrt(d)
|
47 |
attn_weight = q.matmul(k) * softmax_scale
|
48 |
if attn_bias is not None:
|
49 |
+
if (
|
50 |
+
attn_bias.size(-1) != 1
|
51 |
+
and attn_bias.size(-1) != s_k
|
52 |
+
or (attn_bias.size(-2) != 1 and attn_bias.size(-2) != s_q)
|
53 |
+
):
|
54 |
+
raise RuntimeError(
|
55 |
+
f"attn_bias (shape: {attn_bias.shape}) is expected to broadcast to shape: {attn_weight.shape}."
|
56 |
+
)
|
57 |
attn_weight = attn_weight + attn_bias
|
58 |
if key_padding_mask is not None:
|
59 |
if attn_bias is not None:
|
60 |
+
warnings.warn(
|
61 |
+
"Propogating key_padding_mask to the attention module "
|
62 |
+
+ "and applying it within the attention module can cause "
|
63 |
+
+ "unneccessary computation/memory usage. Consider integrating "
|
64 |
+
+ "into attn_bias once and passing that to each attention "
|
65 |
+
+ "module instead."
|
66 |
+
)
|
67 |
+
attn_weight = attn_weight.masked_fill(
|
68 |
+
~key_padding_mask.view((b, 1, 1, s_k)), min_val
|
69 |
+
)
|
70 |
if is_causal:
|
71 |
s = max(s_q, s_k)
|
72 |
causal_mask = attn_weight.new_ones(s, s, dtype=torch.float16)
|
|
|
77 |
attn_weight = attn_weight.masked_fill(causal_mask.view(1, 1, s_q, s_k), min_val)
|
78 |
attn_weight = torch.softmax(attn_weight, dim=-1)
|
79 |
if dropout_p:
|
80 |
+
attn_weight = torch.nn.functional.dropout(
|
81 |
+
attn_weight, p=dropout_p, training=training, inplace=True
|
82 |
+
)
|
83 |
out = attn_weight.matmul(v)
|
84 |
+
out = rearrange(out, "b h s d -> b s (h d)")
|
85 |
if needs_weights:
|
86 |
return (out, attn_weight)
|
87 |
return (out, None)
|
88 |
|
89 |
+
|
90 |
def check_valid_inputs(*tensors, valid_dtypes=[torch.float16, torch.bfloat16]):
|
91 |
for tensor in tensors:
|
92 |
if tensor.dtype not in valid_dtypes:
|
93 |
+
raise TypeError(
|
94 |
+
f"tensor.dtype={tensor.dtype!r} must be in valid_dtypes={valid_dtypes!r}."
|
95 |
+
)
|
96 |
if not tensor.is_cuda:
|
97 |
+
raise TypeError(
|
98 |
+
f"Inputs must be cuda tensors (tensor.is_cuda={tensor.is_cuda!r})."
|
99 |
+
)
|
100 |
+
|
101 |
|
102 |
+
def flash_attn_fn(
|
103 |
+
query,
|
104 |
+
key,
|
105 |
+
value,
|
106 |
+
n_heads,
|
107 |
+
softmax_scale=None,
|
108 |
+
attn_bias=None,
|
109 |
+
key_padding_mask=None,
|
110 |
+
is_causal=False,
|
111 |
+
dropout_p=0.0,
|
112 |
+
training=False,
|
113 |
+
needs_weights=False,
|
114 |
+
multiquery=False,
|
115 |
+
):
|
116 |
try:
|
117 |
from flash_attn import bert_padding, flash_attn_interface
|
118 |
except:
|
119 |
+
raise RuntimeError("Please install flash-attn==1.0.3.post0")
|
120 |
check_valid_inputs(query, key, value)
|
121 |
if attn_bias is not None:
|
122 |
+
raise NotImplementedError(f"attn_bias not implemented for flash attn.")
|
123 |
(batch_size, seqlen) = query.shape[:2]
|
124 |
if key_padding_mask is None:
|
125 |
key_padding_mask = torch.ones_like(key[:, :, 0], dtype=torch.bool)
|
126 |
+
query_padding_mask = key_padding_mask[:, -query.size(1) :]
|
127 |
+
(query_unpad, indices_q, cu_seqlens_q, max_seqlen_q) = bert_padding.unpad_input(
|
128 |
+
query, query_padding_mask
|
129 |
+
)
|
130 |
+
query_unpad = rearrange(query_unpad, "nnz (h d) -> nnz h d", h=n_heads)
|
131 |
+
(key_unpad, _, cu_seqlens_k, max_seqlen_k) = bert_padding.unpad_input(
|
132 |
+
key, key_padding_mask
|
133 |
+
)
|
134 |
+
key_unpad = rearrange(
|
135 |
+
key_unpad, "nnz (h d) -> nnz h d", h=1 if multiquery else n_heads
|
136 |
+
)
|
137 |
(value_unpad, _, _, _) = bert_padding.unpad_input(value, key_padding_mask)
|
138 |
+
value_unpad = rearrange(
|
139 |
+
value_unpad, "nnz (h d) -> nnz h d", h=1 if multiquery else n_heads
|
140 |
+
)
|
141 |
if multiquery:
|
142 |
key_unpad = key_unpad.expand(key_unpad.size(0), n_heads, key_unpad.size(-1))
|
143 |
+
value_unpad = value_unpad.expand(
|
144 |
+
value_unpad.size(0), n_heads, value_unpad.size(-1)
|
145 |
+
)
|
146 |
dropout_p = dropout_p if training else 0.0
|
147 |
reset_is_causal = _reset_is_causal(query.size(1), key.size(1), is_causal)
|
148 |
+
output_unpad = flash_attn_interface.flash_attn_unpadded_func(
|
149 |
+
query_unpad,
|
150 |
+
key_unpad,
|
151 |
+
value_unpad,
|
152 |
+
cu_seqlens_q,
|
153 |
+
cu_seqlens_k,
|
154 |
+
max_seqlen_q,
|
155 |
+
max_seqlen_k,
|
156 |
+
dropout_p,
|
157 |
+
softmax_scale=softmax_scale,
|
158 |
+
causal=reset_is_causal,
|
159 |
+
return_attn_probs=needs_weights,
|
160 |
+
)
|
161 |
+
output = bert_padding.pad_input(
|
162 |
+
rearrange(output_unpad, "nnz h d -> nnz (h d)"), indices_q, batch_size, seqlen
|
163 |
+
)
|
164 |
return (output, None)
|
165 |
|
166 |
+
|
167 |
+
def triton_flash_attn_fn(
|
168 |
+
query,
|
169 |
+
key,
|
170 |
+
value,
|
171 |
+
n_heads,
|
172 |
+
softmax_scale=None,
|
173 |
+
attn_bias=None,
|
174 |
+
key_padding_mask=None,
|
175 |
+
is_causal=False,
|
176 |
+
dropout_p=0.0,
|
177 |
+
training=False,
|
178 |
+
needs_weights=False,
|
179 |
+
multiquery=False,
|
180 |
+
):
|
181 |
try:
|
182 |
from flash_attn import flash_attn_triton
|
183 |
except:
|
184 |
+
raise RuntimeError(
|
185 |
+
"Please install flash-attn==1.0.3.post0 and triton==2.0.0.dev20221202"
|
186 |
+
)
|
187 |
check_valid_inputs(query, key, value)
|
188 |
if dropout_p:
|
189 |
+
raise NotImplementedError(f"Dropout not implemented for attn_impl: triton.")
|
190 |
if needs_weights:
|
191 |
+
raise NotImplementedError(f"attn_impl: triton cannot return attn weights.")
|
192 |
if key_padding_mask is not None:
|
193 |
+
warnings.warn(
|
194 |
+
"Propagating key_padding_mask to the attention module "
|
195 |
+
+ "and applying it within the attention module can cause "
|
196 |
+
+ "unnecessary computation/memory usage. Consider integrating "
|
197 |
+
+ "into attn_bias once and passing that to each attention "
|
198 |
+
+ "module instead."
|
199 |
+
)
|
200 |
(b_size, s_k) = key_padding_mask.shape[:2]
|
201 |
if attn_bias is None:
|
202 |
attn_bias = query.new_zeros(b_size, 1, 1, s_k)
|
203 |
+
attn_bias = attn_bias.masked_fill(
|
204 |
+
~key_padding_mask.view((b_size, 1, 1, s_k)), torch.finfo(query.dtype).min
|
205 |
+
)
|
206 |
+
query = rearrange(query, "b s (h d) -> b s h d", h=n_heads)
|
207 |
+
key = rearrange(key, "b s (h d) -> b s h d", h=1 if multiquery else n_heads)
|
208 |
+
value = rearrange(value, "b s (h d) -> b s h d", h=1 if multiquery else n_heads)
|
209 |
if multiquery:
|
210 |
key = key.expand(*key.shape[:2], n_heads, key.size(-1))
|
211 |
value = value.expand(*value.shape[:2], n_heads, value.size(-1))
|
212 |
reset_is_causal = _reset_is_causal(query.size(1), key.size(1), is_causal)
|
213 |
+
attn_output = flash_attn_triton.flash_attn_func(
|
214 |
+
query, key, value, attn_bias, reset_is_causal, softmax_scale
|
215 |
+
)
|
216 |
output = attn_output.view(*attn_output.shape[:2], -1)
|
217 |
return (output, None)
|
218 |
|
219 |
+
|
220 |
class MultiheadAttention(nn.Module):
|
221 |
"""Multi-head self attention.
|
222 |
|
|
|
224 |
additive bias.
|
225 |
"""
|
226 |
|
227 |
+
def __init__(
|
228 |
+
self,
|
229 |
+
d_model: int,
|
230 |
+
n_heads: int,
|
231 |
+
attn_impl: str = "triton",
|
232 |
+
clip_qkv: Optional[float] = None,
|
233 |
+
qk_ln: bool = False,
|
234 |
+
softmax_scale: Optional[float] = None,
|
235 |
+
attn_pdrop: float = 0.0,
|
236 |
+
low_precision_layernorm: bool = False,
|
237 |
+
device: Optional[str] = None,
|
238 |
+
):
|
239 |
super().__init__()
|
240 |
self.attn_impl = attn_impl
|
241 |
self.clip_qkv = clip_qkv
|
|
|
253 |
layernorm_class = LPLayerNorm if low_precision_layernorm else nn.LayerNorm
|
254 |
self.q_ln = layernorm_class(self.d_model, device=device)
|
255 |
self.k_ln = layernorm_class(self.d_model, device=device)
|
256 |
+
if self.attn_impl == "flash":
|
257 |
self.attn_fn = flash_attn_fn
|
258 |
+
elif self.attn_impl == "triton":
|
259 |
self.attn_fn = triton_flash_attn_fn
|
260 |
+
warnings.warn(
|
261 |
+
"While `attn_impl: triton` can be faster than `attn_impl: flash` "
|
262 |
+
+ "it uses more memory. When training larger models this can trigger "
|
263 |
+
+ "alloc retries which hurts performance. If encountered, we recommend "
|
264 |
+
+ "using `attn_impl: flash` if your model does not use `alibi` or `prefix_lm`."
|
265 |
+
)
|
266 |
+
elif self.attn_impl == "torch":
|
267 |
self.attn_fn = scaled_multihead_dot_product_attention
|
268 |
if torch.cuda.is_available():
|
269 |
+
warnings.warn(
|
270 |
+
"Using `attn_impl: torch`. If your model does not use `alibi` or "
|
271 |
+
+ "`prefix_lm` we recommend using `attn_impl: flash` otherwise "
|
272 |
+
+ "we recommend using `attn_impl: triton`."
|
273 |
+
)
|
274 |
else:
|
275 |
+
raise ValueError(f"attn_impl={attn_impl!r} is an invalid setting.")
|
276 |
self.out_proj = nn.Linear(self.d_model, self.d_model, device=device)
|
277 |
self.out_proj._is_residual = True
|
278 |
|
279 |
+
def forward(
|
280 |
+
self,
|
281 |
+
x,
|
282 |
+
past_key_value=None,
|
283 |
+
attn_bias=None,
|
284 |
+
attention_mask=None,
|
285 |
+
is_causal=True,
|
286 |
+
needs_weights=False,
|
287 |
+
):
|
288 |
qkv = self.Wqkv(x)
|
289 |
if self.clip_qkv:
|
290 |
qkv.clamp_(min=-self.clip_qkv, max=self.clip_qkv)
|
|
|
300 |
value = torch.cat([past_key_value[1], value], dim=1)
|
301 |
past_key_value = (key, value)
|
302 |
if attn_bias is not None:
|
303 |
+
attn_bias = attn_bias[:, :, -query.size(1) :, -key.size(1) :]
|
304 |
+
(context, attn_weights) = self.attn_fn(
|
305 |
+
query,
|
306 |
+
key,
|
307 |
+
value,
|
308 |
+
self.n_heads,
|
309 |
+
softmax_scale=self.softmax_scale,
|
310 |
+
attn_bias=attn_bias,
|
311 |
+
key_padding_mask=key_padding_mask,
|
312 |
+
is_causal=is_causal,
|
313 |
+
dropout_p=self.attn_dropout_p,
|
314 |
+
training=self.training,
|
315 |
+
needs_weights=needs_weights,
|
316 |
+
)
|
317 |
return (self.out_proj(context), attn_weights, past_key_value)
|
318 |
|
319 |
+
|
320 |
class MultiQueryAttention(nn.Module):
|
321 |
"""Multi-Query self attention.
|
322 |
|
|
|
324 |
additive bias.
|
325 |
"""
|
326 |
|
327 |
+
def __init__(
|
328 |
+
self,
|
329 |
+
d_model: int,
|
330 |
+
n_heads: int,
|
331 |
+
attn_impl: str = "triton",
|
332 |
+
clip_qkv: Optional[float] = None,
|
333 |
+
qk_ln: bool = False,
|
334 |
+
softmax_scale: Optional[float] = None,
|
335 |
+
attn_pdrop: float = 0.0,
|
336 |
+
low_precision_layernorm: bool = False,
|
337 |
+
device: Optional[str] = None,
|
338 |
+
):
|
339 |
super().__init__()
|
340 |
self.attn_impl = attn_impl
|
341 |
self.clip_qkv = clip_qkv
|
|
|
354 |
layernorm_class = LPLayerNorm if low_precision_layernorm else nn.LayerNorm
|
355 |
self.q_ln = layernorm_class(d_model, device=device)
|
356 |
self.k_ln = layernorm_class(self.head_dim, device=device)
|
357 |
+
if self.attn_impl == "flash":
|
358 |
self.attn_fn = flash_attn_fn
|
359 |
+
elif self.attn_impl == "triton":
|
360 |
self.attn_fn = triton_flash_attn_fn
|
361 |
+
warnings.warn(
|
362 |
+
"While `attn_impl: triton` can be faster than `attn_impl: flash` "
|
363 |
+
+ "it uses more memory. When training larger models this can trigger "
|
364 |
+
+ "alloc retries which hurts performance. If encountered, we recommend "
|
365 |
+
+ "using `attn_impl: flash` if your model does not use `alibi` or `prefix_lm`."
|
366 |
+
)
|
367 |
+
elif self.attn_impl == "torch":
|
368 |
self.attn_fn = scaled_multihead_dot_product_attention
|
369 |
if torch.cuda.is_available():
|
370 |
+
warnings.warn(
|
371 |
+
"Using `attn_impl: torch`. If your model does not use `alibi` or "
|
372 |
+
+ "`prefix_lm` we recommend using `attn_impl: flash` otherwise "
|
373 |
+
+ "we recommend using `attn_impl: triton`."
|
374 |
+
)
|
375 |
else:
|
376 |
+
raise ValueError(f"attn_impl={attn_impl!r} is an invalid setting.")
|
377 |
self.out_proj = nn.Linear(self.d_model, self.d_model, device=device)
|
378 |
self.out_proj._is_residual = True
|
379 |
|
380 |
+
def forward(
|
381 |
+
self,
|
382 |
+
x,
|
383 |
+
past_key_value=None,
|
384 |
+
attn_bias=None,
|
385 |
+
attention_mask=None,
|
386 |
+
is_causal=True,
|
387 |
+
needs_weights=False,
|
388 |
+
):
|
389 |
qkv = self.Wqkv(x)
|
390 |
if self.clip_qkv:
|
391 |
qkv.clamp_(min=-self.clip_qkv, max=self.clip_qkv)
|
392 |
+
(query, key, value) = qkv.split(
|
393 |
+
[self.d_model, self.head_dim, self.head_dim], dim=2
|
394 |
+
)
|
395 |
key_padding_mask = attention_mask
|
396 |
if self.qk_ln:
|
397 |
dtype = query.dtype
|
|
|
403 |
value = torch.cat([past_key_value[1], value], dim=1)
|
404 |
past_key_value = (key, value)
|
405 |
if attn_bias is not None:
|
406 |
+
attn_bias = attn_bias[:, :, -query.size(1) :, -key.size(1) :]
|
407 |
+
(context, attn_weights) = self.attn_fn(
|
408 |
+
query,
|
409 |
+
key,
|
410 |
+
value,
|
411 |
+
self.n_heads,
|
412 |
+
softmax_scale=self.softmax_scale,
|
413 |
+
attn_bias=attn_bias,
|
414 |
+
key_padding_mask=key_padding_mask,
|
415 |
+
is_causal=is_causal,
|
416 |
+
dropout_p=self.attn_dropout_p,
|
417 |
+
training=self.training,
|
418 |
+
needs_weights=needs_weights,
|
419 |
+
multiquery=True,
|
420 |
+
)
|
421 |
return (self.out_proj(context), attn_weights, past_key_value)
|
422 |
|
423 |
+
|
424 |
+
def attn_bias_shape(
|
425 |
+
attn_impl, n_heads, seq_len, alibi, prefix_lm, causal, use_sequence_id
|
426 |
+
):
|
427 |
+
if attn_impl == "flash":
|
428 |
return None
|
429 |
+
elif attn_impl in ["torch", "triton"]:
|
430 |
if alibi:
|
431 |
if (prefix_lm or not causal) or use_sequence_id:
|
432 |
return (1, n_heads, seq_len, seq_len)
|
|
|
435 |
return (1, 1, seq_len, seq_len)
|
436 |
return None
|
437 |
else:
|
438 |
+
raise ValueError(f"attn_impl={attn_impl!r} is an invalid setting.")
|
439 |
+
|
440 |
|
441 |
+
def build_attn_bias(
|
442 |
+
attn_impl, attn_bias, n_heads, seq_len, causal=False, alibi=False, alibi_bias_max=8
|
443 |
+
):
|
444 |
+
if attn_impl == "flash":
|
445 |
return None
|
446 |
+
elif attn_impl in ["torch", "triton"]:
|
447 |
if alibi:
|
448 |
(device, dtype) = (attn_bias.device, attn_bias.dtype)
|
449 |
+
attn_bias = attn_bias.add(
|
450 |
+
build_alibi_bias(
|
451 |
+
n_heads,
|
452 |
+
seq_len,
|
453 |
+
full=not causal,
|
454 |
+
alibi_bias_max=alibi_bias_max,
|
455 |
+
device=device,
|
456 |
+
dtype=dtype,
|
457 |
+
)
|
458 |
+
)
|
459 |
return attn_bias
|
460 |
else:
|
461 |
+
raise ValueError(f"attn_impl={attn_impl!r} is an invalid setting.")
|
462 |
+
|
463 |
|
464 |
def gen_slopes(n_heads, alibi_bias_max=8, device=None):
|
465 |
_n_heads = 2 ** math.ceil(math.log2(n_heads))
|
|
|
470 |
slopes = torch.concat([slopes[1::2], slopes[::2]])[:n_heads]
|
471 |
return slopes.view(1, n_heads, 1, 1)
|
472 |
|
473 |
+
|
474 |
+
def build_alibi_bias(
|
475 |
+
n_heads, seq_len, full=False, alibi_bias_max=8, device=None, dtype=None
|
476 |
+
):
|
477 |
+
alibi_bias = torch.arange(1 - seq_len, 1, dtype=torch.int32, device=device).view(
|
478 |
+
1, 1, 1, seq_len
|
479 |
+
)
|
480 |
if full:
|
481 |
+
alibi_bias = alibi_bias - torch.arange(
|
482 |
+
1 - seq_len, 1, dtype=torch.int32, device=device
|
483 |
+
).view(1, 1, seq_len, 1)
|
484 |
alibi_bias = alibi_bias.abs().mul(-1)
|
485 |
slopes = gen_slopes(n_heads, alibi_bias_max, device=device)
|
486 |
alibi_bias = alibi_bias * slopes
|
487 |
return alibi_bias.to(dtype=dtype)
|
488 |
+
|
489 |
+
|
490 |
+
ATTN_CLASS_REGISTRY = {
|
491 |
+
"multihead_attention": MultiheadAttention,
|
492 |
+
"multiquery_attention": MultiQueryAttention,
|
493 |
+
}
|
blocks.py
CHANGED
@@ -5,37 +5,84 @@ import torch.nn as nn
|
|
5 |
from .attention import ATTN_CLASS_REGISTRY
|
6 |
from .norm import NORM_CLASS_REGISTRY
|
7 |
|
8 |
-
class MPTMLP(nn.Module):
|
9 |
|
10 |
-
|
|
|
|
|
|
|
11 |
super().__init__()
|
12 |
self.up_proj = nn.Linear(d_model, expansion_ratio * d_model, device=device)
|
13 |
-
self.act = nn.GELU(approximate=
|
14 |
self.down_proj = nn.Linear(expansion_ratio * d_model, d_model, device=device)
|
15 |
self.down_proj._is_residual = True
|
16 |
|
17 |
def forward(self, x):
|
18 |
return self.down_proj(self.act(self.up_proj(x)))
|
19 |
|
20 |
-
class MPTBlock(nn.Module):
|
21 |
|
22 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
23 |
del kwargs
|
24 |
super().__init__()
|
25 |
norm_class = NORM_CLASS_REGISTRY[norm_type.lower()]
|
26 |
-
attn_class = ATTN_CLASS_REGISTRY[attn_config[
|
27 |
self.norm_1 = norm_class(d_model, device=device)
|
28 |
-
self.attn = attn_class(
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
29 |
self.norm_2 = norm_class(d_model, device=device)
|
30 |
-
self.ffn = MPTMLP(
|
|
|
|
|
31 |
self.resid_attn_dropout = nn.Dropout(resid_pdrop)
|
32 |
self.resid_ffn_dropout = nn.Dropout(resid_pdrop)
|
33 |
|
34 |
-
def forward(
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
35 |
a = self.norm_1(x)
|
36 |
-
(b, _, past_key_value) = self.attn(
|
|
|
|
|
|
|
|
|
|
|
|
|
37 |
x = x + self.resid_attn_dropout(b)
|
38 |
m = self.norm_2(x)
|
39 |
n = self.ffn(m)
|
40 |
x = x + self.resid_ffn_dropout(n)
|
41 |
-
return (x, past_key_value)
|
|
|
5 |
from .attention import ATTN_CLASS_REGISTRY
|
6 |
from .norm import NORM_CLASS_REGISTRY
|
7 |
|
|
|
8 |
|
9 |
+
class MPTMLP(nn.Module):
|
10 |
+
def __init__(
|
11 |
+
self, d_model: int, expansion_ratio: int, device: Optional[str] = None
|
12 |
+
):
|
13 |
super().__init__()
|
14 |
self.up_proj = nn.Linear(d_model, expansion_ratio * d_model, device=device)
|
15 |
+
self.act = nn.GELU(approximate="none")
|
16 |
self.down_proj = nn.Linear(expansion_ratio * d_model, d_model, device=device)
|
17 |
self.down_proj._is_residual = True
|
18 |
|
19 |
def forward(self, x):
|
20 |
return self.down_proj(self.act(self.up_proj(x)))
|
21 |
|
|
|
22 |
|
23 |
+
class MPTBlock(nn.Module):
|
24 |
+
def __init__(
|
25 |
+
self,
|
26 |
+
d_model: int,
|
27 |
+
n_heads: int,
|
28 |
+
expansion_ratio: int,
|
29 |
+
attn_config: Dict = {
|
30 |
+
"attn_type": "multihead_attention",
|
31 |
+
"attn_pdrop": 0.0,
|
32 |
+
"attn_impl": "triton",
|
33 |
+
"qk_ln": False,
|
34 |
+
"clip_qkv": None,
|
35 |
+
"softmax_scale": None,
|
36 |
+
"prefix_lm": False,
|
37 |
+
"attn_uses_sequence_id": False,
|
38 |
+
"alibi": False,
|
39 |
+
"alibi_bias_max": 8,
|
40 |
+
},
|
41 |
+
resid_pdrop: float = 0.0,
|
42 |
+
norm_type: str = "low_precision_layernorm",
|
43 |
+
device: Optional[str] = None,
|
44 |
+
**kwargs
|
45 |
+
):
|
46 |
del kwargs
|
47 |
super().__init__()
|
48 |
norm_class = NORM_CLASS_REGISTRY[norm_type.lower()]
|
49 |
+
attn_class = ATTN_CLASS_REGISTRY[attn_config["attn_type"]]
|
50 |
self.norm_1 = norm_class(d_model, device=device)
|
51 |
+
self.attn = attn_class(
|
52 |
+
attn_impl=attn_config["attn_impl"],
|
53 |
+
clip_qkv=attn_config["clip_qkv"],
|
54 |
+
qk_ln=attn_config["qk_ln"],
|
55 |
+
softmax_scale=attn_config["softmax_scale"],
|
56 |
+
attn_pdrop=attn_config["attn_pdrop"],
|
57 |
+
d_model=d_model,
|
58 |
+
n_heads=n_heads,
|
59 |
+
device=device,
|
60 |
+
)
|
61 |
self.norm_2 = norm_class(d_model, device=device)
|
62 |
+
self.ffn = MPTMLP(
|
63 |
+
d_model=d_model, expansion_ratio=expansion_ratio, device=device
|
64 |
+
)
|
65 |
self.resid_attn_dropout = nn.Dropout(resid_pdrop)
|
66 |
self.resid_ffn_dropout = nn.Dropout(resid_pdrop)
|
67 |
|
68 |
+
def forward(
|
69 |
+
self,
|
70 |
+
x: torch.Tensor,
|
71 |
+
past_key_value: Optional[Tuple[torch.Tensor]] = None,
|
72 |
+
attn_bias: Optional[torch.Tensor] = None,
|
73 |
+
attention_mask: Optional[torch.ByteTensor] = None,
|
74 |
+
is_causal: bool = True,
|
75 |
+
) -> Tuple[torch.Tensor, Optional[Tuple[torch.Tensor]]]:
|
76 |
a = self.norm_1(x)
|
77 |
+
(b, _, past_key_value) = self.attn(
|
78 |
+
a,
|
79 |
+
past_key_value=past_key_value,
|
80 |
+
attn_bias=attn_bias,
|
81 |
+
attention_mask=attention_mask,
|
82 |
+
is_causal=is_causal,
|
83 |
+
)
|
84 |
x = x + self.resid_attn_dropout(b)
|
85 |
m = self.norm_2(x)
|
86 |
n = self.ffn(m)
|
87 |
x = x + self.resid_ffn_dropout(n)
|
88 |
+
return (x, past_key_value)
|
configuration_mpt.py
CHANGED
@@ -1,13 +1,51 @@
|
|
1 |
"""A HuggingFace-style model configuration."""
|
2 |
from typing import Dict, Optional, Union
|
3 |
from transformers import PretrainedConfig
|
4 |
-
|
5 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
6 |
|
7 |
class MPTConfig(PretrainedConfig):
|
8 |
-
model_type =
|
9 |
|
10 |
-
def __init__(
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
11 |
"""The MPT configuration class.
|
12 |
|
13 |
Args:
|
@@ -80,39 +118,76 @@ class MPTConfig(PretrainedConfig):
|
|
80 |
self.norm_type = norm_type
|
81 |
self.use_cache = use_cache
|
82 |
self.init_config = init_config
|
83 |
-
if
|
84 |
-
del kwargs[
|
85 |
-
if
|
86 |
-
del kwargs[
|
87 |
super().__init__(**kwargs)
|
88 |
self._validate_config()
|
89 |
|
90 |
def _set_config_defaults(self, config, config_defaults):
|
91 |
-
for
|
92 |
if k not in config:
|
93 |
config[k] = v
|
94 |
return config
|
95 |
|
96 |
def _validate_config(self):
|
97 |
-
self.attn_config = self._set_config_defaults(
|
98 |
-
|
|
|
|
|
|
|
|
|
99 |
if self.d_model % self.n_heads != 0:
|
100 |
-
raise ValueError(
|
101 |
-
if any(
|
102 |
-
|
103 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
104 |
raise ValueError(f"Unknown attn_impl={self.attn_config['attn_impl']}")
|
105 |
-
if self.attn_config[
|
106 |
-
|
107 |
-
|
108 |
-
|
109 |
-
|
110 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
111 |
if self.embedding_fraction > 1 or self.embedding_fraction <= 0:
|
112 |
-
raise ValueError(
|
113 |
-
|
114 |
-
|
115 |
-
if self.
|
116 |
-
raise ValueError(
|
117 |
-
|
118 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
"""A HuggingFace-style model configuration."""
|
2 |
from typing import Dict, Optional, Union
|
3 |
from transformers import PretrainedConfig
|
4 |
+
|
5 |
+
attn_config_defaults: Dict = {
|
6 |
+
"attn_type": "multihead_attention",
|
7 |
+
"attn_pdrop": 0.0,
|
8 |
+
"attn_impl": "triton",
|
9 |
+
"qk_ln": False,
|
10 |
+
"clip_qkv": None,
|
11 |
+
"softmax_scale": None,
|
12 |
+
"prefix_lm": False,
|
13 |
+
"attn_uses_sequence_id": False,
|
14 |
+
"alibi": False,
|
15 |
+
"alibi_bias_max": 8,
|
16 |
+
}
|
17 |
+
init_config_defaults: Dict = {
|
18 |
+
"name": "kaiming_normal_",
|
19 |
+
"fan_mode": "fan_in",
|
20 |
+
"init_nonlinearity": "relu",
|
21 |
+
}
|
22 |
+
|
23 |
|
24 |
class MPTConfig(PretrainedConfig):
|
25 |
+
model_type = "mpt"
|
26 |
|
27 |
+
def __init__(
|
28 |
+
self,
|
29 |
+
d_model: int = 2048,
|
30 |
+
n_heads: int = 16,
|
31 |
+
n_layers: int = 24,
|
32 |
+
expansion_ratio: int = 4,
|
33 |
+
max_seq_len: int = 2048,
|
34 |
+
vocab_size: int = 50368,
|
35 |
+
resid_pdrop: float = 0.0,
|
36 |
+
emb_pdrop: float = 0.0,
|
37 |
+
learned_pos_emb: bool = True,
|
38 |
+
attn_config: Dict = attn_config_defaults,
|
39 |
+
init_device: str = "cpu",
|
40 |
+
logit_scale: Optional[Union[float, str]] = None,
|
41 |
+
no_bias: bool = False,
|
42 |
+
verbose: int = 0,
|
43 |
+
embedding_fraction: float = 1.0,
|
44 |
+
norm_type: str = "low_precision_layernorm",
|
45 |
+
use_cache: bool = False,
|
46 |
+
init_config: Dict = init_config_defaults,
|
47 |
+
**kwargs,
|
48 |
+
):
|
49 |
"""The MPT configuration class.
|
50 |
|
51 |
Args:
|
|
|
118 |
self.norm_type = norm_type
|
119 |
self.use_cache = use_cache
|
120 |
self.init_config = init_config
|
121 |
+
if "name" in kwargs:
|
122 |
+
del kwargs["name"]
|
123 |
+
if "loss_fn" in kwargs:
|
124 |
+
del kwargs["loss_fn"]
|
125 |
super().__init__(**kwargs)
|
126 |
self._validate_config()
|
127 |
|
128 |
def _set_config_defaults(self, config, config_defaults):
|
129 |
+
for k, v in config_defaults.items():
|
130 |
if k not in config:
|
131 |
config[k] = v
|
132 |
return config
|
133 |
|
134 |
def _validate_config(self):
|
135 |
+
self.attn_config = self._set_config_defaults(
|
136 |
+
self.attn_config, attn_config_defaults
|
137 |
+
)
|
138 |
+
self.init_config = self._set_config_defaults(
|
139 |
+
self.init_config, init_config_defaults
|
140 |
+
)
|
141 |
if self.d_model % self.n_heads != 0:
|
142 |
+
raise ValueError("d_model must be divisible by n_heads")
|
143 |
+
if any(
|
144 |
+
(
|
145 |
+
prob < 0 or prob > 1
|
146 |
+
for prob in [
|
147 |
+
self.attn_config["attn_pdrop"],
|
148 |
+
self.resid_pdrop,
|
149 |
+
self.emb_pdrop,
|
150 |
+
]
|
151 |
+
)
|
152 |
+
):
|
153 |
+
raise ValueError(
|
154 |
+
"self.attn_config['attn_pdrop'], resid_pdrop, emb_pdrop are probabilities and must be between 0 and 1"
|
155 |
+
)
|
156 |
+
if self.attn_config["attn_impl"] not in ["torch", "flash", "triton"]:
|
157 |
raise ValueError(f"Unknown attn_impl={self.attn_config['attn_impl']}")
|
158 |
+
if self.attn_config["prefix_lm"] and self.attn_config["attn_impl"] not in [
|
159 |
+
"torch",
|
160 |
+
"triton",
|
161 |
+
]:
|
162 |
+
raise NotImplementedError(
|
163 |
+
"prefix_lm only implemented with torch and triton attention."
|
164 |
+
)
|
165 |
+
if self.attn_config["alibi"] and self.attn_config["attn_impl"] not in [
|
166 |
+
"torch",
|
167 |
+
"triton",
|
168 |
+
]:
|
169 |
+
raise NotImplementedError(
|
170 |
+
"alibi only implemented with torch and triton attention."
|
171 |
+
)
|
172 |
+
if self.attn_config["attn_uses_sequence_id"] and self.attn_config[
|
173 |
+
"attn_impl"
|
174 |
+
] not in ["torch", "triton"]:
|
175 |
+
raise NotImplementedError(
|
176 |
+
"attn_uses_sequence_id only implemented with torch and triton attention."
|
177 |
+
)
|
178 |
if self.embedding_fraction > 1 or self.embedding_fraction <= 0:
|
179 |
+
raise ValueError(
|
180 |
+
"model.embedding_fraction must be between 0 (exclusive) and 1 (inclusive)!"
|
181 |
+
)
|
182 |
+
if isinstance(self.logit_scale, str) and self.logit_scale != "inv_sqrt_d_model":
|
183 |
+
raise ValueError(
|
184 |
+
f"self.logit_scale={self.logit_scale!r} is not recognized as an option; use numeric value or 'inv_sqrt_d_model'."
|
185 |
+
)
|
186 |
+
if self.init_config.get("name", None) is None:
|
187 |
+
raise ValueError(
|
188 |
+
f"self.init_config={self.init_config!r} 'name' needs to be set."
|
189 |
+
)
|
190 |
+
if not self.learned_pos_emb and (not self.attn_config["alibi"]):
|
191 |
+
raise ValueError(
|
192 |
+
f"Positional information must be provided to the model using either learned_pos_emb or alibi."
|
193 |
+
)
|
hf_prefixlm_converter.py
CHANGED
@@ -11,9 +11,17 @@ import warnings
|
|
11 |
from types import MethodType
|
12 |
from typing import Any, Dict, List, Optional, Tuple, Union
|
13 |
import torch
|
14 |
-
from transformers.models.bloom.modeling_bloom import
|
|
|
|
|
|
|
|
|
|
|
|
|
15 |
from transformers.models.bloom.modeling_bloom import _expand_mask as _expand_mask_bloom
|
16 |
-
from transformers.models.bloom.modeling_bloom import
|
|
|
|
|
17 |
from transformers.models.bloom.modeling_bloom import logging
|
18 |
from transformers.models.gpt2.modeling_gpt2 import GPT2LMHeadModel
|
19 |
from transformers.models.gpt_neo.modeling_gpt_neo import GPTNeoForCausalLM
|
@@ -21,10 +29,21 @@ from transformers.models.gpt_neox.modeling_gpt_neox import GPTNeoXForCausalLM
|
|
21 |
from transformers.models.gptj.modeling_gptj import GPTJForCausalLM
|
22 |
from transformers.models.opt.modeling_opt import OPTForCausalLM
|
23 |
from transformers.models.opt.modeling_opt import _expand_mask as _expand_mask_opt
|
24 |
-
from transformers.models.opt.modeling_opt import
|
|
|
|
|
|
|
25 |
logger = logging.get_logger(__name__)
|
26 |
-
_SUPPORTED_GPT_MODELS = (
|
27 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
28 |
|
29 |
def _convert_gpt_causal_lm_to_prefix_lm(model: CAUSAL_GPT_TYPES) -> CAUSAL_GPT_TYPES:
|
30 |
"""Converts a GPT-style Causal LM to a Prefix LM.
|
@@ -37,10 +56,12 @@ def _convert_gpt_causal_lm_to_prefix_lm(model: CAUSAL_GPT_TYPES) -> CAUSAL_GPT_T
|
|
37 |
|
38 |
See `convert_hf_causal_lm_to_prefix_lm` for more details.
|
39 |
"""
|
40 |
-
if hasattr(model,
|
41 |
return model
|
42 |
assert isinstance(model, _SUPPORTED_GPT_MODELS)
|
43 |
-
assert
|
|
|
|
|
44 |
|
45 |
def _get_attn_modules(model: CAUSAL_GPT_TYPES) -> List[torch.nn.Module]:
|
46 |
"""Helper that gets a list of the model's attention modules.
|
@@ -56,7 +77,7 @@ def _convert_gpt_causal_lm_to_prefix_lm(model: CAUSAL_GPT_TYPES) -> CAUSAL_GPT_T
|
|
56 |
blocks = model.transformer.h
|
57 |
for block in blocks:
|
58 |
if isinstance(model, GPTNeoForCausalLM):
|
59 |
-
if block.attn.attention_type !=
|
60 |
continue
|
61 |
attn_module = block.attn.attention
|
62 |
elif isinstance(model, GPTNeoXForCausalLM):
|
@@ -65,17 +86,58 @@ def _convert_gpt_causal_lm_to_prefix_lm(model: CAUSAL_GPT_TYPES) -> CAUSAL_GPT_T
|
|
65 |
attn_module = block.attn
|
66 |
attn_modules.append(attn_module)
|
67 |
return attn_modules
|
68 |
-
setattr(model, '_original_forward', getattr(model, 'forward'))
|
69 |
-
setattr(model, '_original_generate', getattr(model, 'generate'))
|
70 |
|
71 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
72 |
"""Wraps original forward to enable PrefixLM attention."""
|
73 |
|
74 |
def call_og_forward():
|
75 |
if isinstance(self, GPTNeoXForCausalLM):
|
76 |
-
return self._original_forward(
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
77 |
else:
|
78 |
-
return self._original_forward(
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
79 |
if bidirectional_mask is None:
|
80 |
return call_og_forward()
|
81 |
assert isinstance(bidirectional_mask, torch.Tensor)
|
@@ -83,14 +145,23 @@ def _convert_gpt_causal_lm_to_prefix_lm(model: CAUSAL_GPT_TYPES) -> CAUSAL_GPT_T
|
|
83 |
(b, s) = bidirectional_mask.shape
|
84 |
max_length = attn_modules[0].bias.shape[-1]
|
85 |
if s > max_length:
|
86 |
-
raise ValueError(
|
|
|
|
|
|
|
87 |
assert s <= max_length
|
88 |
if s < max_length:
|
89 |
-
pad = torch.zeros(
|
|
|
|
|
|
|
|
|
90 |
bidirectional_mask = torch.cat([bidirectional_mask, pad], dim=1)
|
91 |
bidirectional = bidirectional_mask.unsqueeze(1).unsqueeze(1)
|
92 |
for attn_module in attn_modules:
|
93 |
-
attn_module.bias.data = torch.logical_or(
|
|
|
|
|
94 |
output = call_og_forward()
|
95 |
for attn_module in attn_modules:
|
96 |
attn_module.bias.data = torch.tril(attn_module.bias.data[0, 0])[None, None]
|
@@ -105,11 +176,13 @@ def _convert_gpt_causal_lm_to_prefix_lm(model: CAUSAL_GPT_TYPES) -> CAUSAL_GPT_T
|
|
105 |
for attn_module in attn_modules:
|
106 |
attn_module.bias.data = torch.tril(attn_module.bias.data[0, 0])[None, None]
|
107 |
return output
|
108 |
-
|
109 |
-
setattr(model,
|
110 |
-
setattr(model,
|
|
|
111 |
return model
|
112 |
|
|
|
113 |
def _convert_bloom_causal_lm_to_prefix_lm(model: BloomForCausalLM) -> BloomForCausalLM:
|
114 |
"""Converts a BLOOM Causal LM to a Prefix LM.
|
115 |
|
@@ -118,62 +191,137 @@ def _convert_bloom_causal_lm_to_prefix_lm(model: BloomForCausalLM) -> BloomForCa
|
|
118 |
|
119 |
See `convert_hf_causal_lm_to_prefix_lm` for more details.
|
120 |
"""
|
121 |
-
if hasattr(model,
|
122 |
return model
|
123 |
assert isinstance(model, BloomForCausalLM)
|
124 |
-
assert
|
125 |
-
|
126 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
127 |
combined_attention_mask = None
|
128 |
device = attention_mask.device
|
129 |
(_, src_length) = input_shape
|
130 |
if src_length > 1:
|
131 |
-
combined_attention_mask = _make_causal_mask_bloom(
|
|
|
|
|
|
|
|
|
132 |
if bidirectional_mask is not None:
|
133 |
assert attention_mask.shape == bidirectional_mask.shape
|
134 |
-
expanded_bidirectional_mask = _expand_mask_bloom(
|
135 |
-
|
|
|
|
|
|
|
|
|
136 |
expanded_attn_mask = _expand_mask_bloom(attention_mask, tgt_length=src_length)
|
137 |
-
combined_attention_mask =
|
|
|
|
|
|
|
|
|
138 |
return combined_attention_mask
|
139 |
|
140 |
-
def _build_alibi_tensor(
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
141 |
num_heads = self.config.n_head
|
142 |
closest_power_of_2 = 2 ** math.floor(math.log2(num_heads))
|
143 |
-
base = torch.tensor(
|
144 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
145 |
slopes = torch.pow(base, powers)
|
146 |
if closest_power_of_2 != num_heads:
|
147 |
-
extra_base = torch.tensor(
|
148 |
-
|
149 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
150 |
slopes = torch.cat([slopes, torch.pow(extra_base, extra_powers)], dim=0)
|
151 |
qa = torch.arange(query_length, device=device, dtype=torch.int32).view(-1, 1)
|
152 |
ka = torch.arange(key_length, device=device, dtype=torch.int32).view(1, -1)
|
153 |
diffs = qa - ka + key_length - query_length
|
154 |
diffs = -diffs.abs()
|
155 |
-
alibi = slopes.view(1, num_heads, 1, 1) * diffs.view(
|
156 |
-
|
|
|
|
|
|
|
|
|
157 |
return alibi.to(dtype)
|
|
|
158 |
KeyValueT = Tuple[torch.Tensor, torch.Tensor]
|
159 |
|
160 |
-
def forward(
|
161 |
-
|
162 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
163 |
if len(deprecated_arguments) > 0:
|
164 |
-
raise ValueError(f
|
165 |
-
output_attentions =
|
166 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
167 |
use_cache = use_cache if use_cache is not None else self.config.use_cache
|
168 |
-
return_dict =
|
|
|
|
|
169 |
if input_ids is not None and inputs_embeds is not None:
|
170 |
-
raise ValueError(
|
|
|
|
|
171 |
elif input_ids is not None:
|
172 |
(batch_size, seq_length) = input_ids.shape
|
173 |
elif inputs_embeds is not None:
|
174 |
(batch_size, seq_length, _) = inputs_embeds.shape
|
175 |
else:
|
176 |
-
raise ValueError(
|
177 |
if past_key_values is None:
|
178 |
past_key_values = tuple([None] * len(self.h))
|
179 |
head_mask = self.get_head_mask(head_mask, self.config.n_layer)
|
@@ -190,28 +338,62 @@ def _convert_bloom_causal_lm_to_prefix_lm(model: BloomForCausalLM) -> BloomForCa
|
|
190 |
past_key_values_length = tmp.shape[2]
|
191 |
seq_length_with_past = seq_length_with_past + past_key_values_length
|
192 |
if attention_mask is None:
|
193 |
-
attention_mask = torch.ones(
|
|
|
|
|
194 |
else:
|
195 |
attention_mask = attention_mask.to(hidden_states.device)
|
196 |
-
alibi = self._build_alibi_tensor(
|
197 |
-
|
198 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
199 |
if output_hidden_states:
|
200 |
hst = (hidden_states,)
|
201 |
all_hidden_states = all_hidden_states + hst
|
202 |
if self.gradient_checkpointing and self.training:
|
203 |
if use_cache:
|
204 |
-
logger.warning(
|
|
|
|
|
205 |
use_cache = False
|
206 |
|
207 |
def create_custom_forward(module):
|
208 |
-
|
209 |
def custom_forward(*inputs):
|
210 |
-
return module(
|
|
|
|
|
|
|
|
|
|
|
211 |
return custom_forward
|
212 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
213 |
else:
|
214 |
-
outputs = block(
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
215 |
hidden_states = outputs[0]
|
216 |
if use_cache is True:
|
217 |
presents = presents + (outputs[1],)
|
@@ -223,21 +405,77 @@ def _convert_bloom_causal_lm_to_prefix_lm(model: BloomForCausalLM) -> BloomForCa
|
|
223 |
hst = (hidden_states,)
|
224 |
all_hidden_states = all_hidden_states + hst
|
225 |
if not return_dict:
|
226 |
-
return tuple(
|
227 |
-
|
228 |
-
|
229 |
-
|
230 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
231 |
KeyValueT = Tuple[torch.Tensor, torch.Tensor]
|
232 |
|
233 |
-
def forward(
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
234 |
"""Replacement forward method for BloomCausalLM."""
|
235 |
-
if deprecated_arguments.pop(
|
236 |
-
warnings.warn(
|
|
|
|
|
|
|
|
|
237 |
if len(deprecated_arguments) > 0:
|
238 |
-
raise ValueError(f
|
239 |
-
return_dict =
|
240 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
241 |
hidden_states = transformer_outputs[0]
|
242 |
lm_logits = self.lm_head(hidden_states)
|
243 |
loss = None
|
@@ -246,13 +484,28 @@ def _convert_bloom_causal_lm_to_prefix_lm(model: BloomForCausalLM) -> BloomForCa
|
|
246 |
shift_labels = labels[..., 1:].contiguous()
|
247 |
(batch_size, seq_length, vocab_size) = shift_logits.shape
|
248 |
loss_fct = CrossEntropyLoss()
|
249 |
-
loss = loss_fct(
|
|
|
|
|
|
|
250 |
if not return_dict:
|
251 |
output = (lm_logits,) + transformer_outputs[1:]
|
252 |
return (loss,) + output if loss is not None else output
|
253 |
-
return CausalLMOutputWithCrossAttentions(
|
254 |
-
|
255 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
256 |
if past:
|
257 |
input_ids = input_ids[:, -1].unsqueeze(-1)
|
258 |
bidirectional_mask = None
|
@@ -260,12 +513,24 @@ def _convert_bloom_causal_lm_to_prefix_lm(model: BloomForCausalLM) -> BloomForCa
|
|
260 |
past = self._convert_to_bloom_cache(past)
|
261 |
else:
|
262 |
bidirectional_mask = torch.ones_like(input_ids)
|
263 |
-
return {
|
264 |
-
|
265 |
-
|
266 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
267 |
return model
|
268 |
|
|
|
269 |
def _convert_opt_causal_lm_to_prefix_lm(model: OPTForCausalLM) -> OPTForCausalLM:
|
270 |
"""Converts an OPT Causal LM to a Prefix LM.
|
271 |
|
@@ -274,36 +539,89 @@ def _convert_opt_causal_lm_to_prefix_lm(model: OPTForCausalLM) -> OPTForCausalLM
|
|
274 |
|
275 |
See `convert_hf_causal_lm_to_prefix_lm` for more details.
|
276 |
"""
|
277 |
-
if hasattr(model,
|
278 |
return model
|
279 |
assert isinstance(model, OPTForCausalLM)
|
280 |
-
assert
|
281 |
-
|
282 |
-
|
|
|
|
|
283 |
model.model.decoder.bidirectional_mask = None
|
284 |
|
285 |
-
def _prepare_decoder_attention_mask(
|
|
|
|
|
286 |
combined_attention_mask = None
|
287 |
if input_shape[-1] > 1:
|
288 |
-
if self.bidirectional_mask ==
|
289 |
(bsz, src_length) = input_shape
|
290 |
-
combined_attention_mask = torch.zeros(
|
|
|
|
|
|
|
|
|
291 |
else:
|
292 |
-
combined_attention_mask = _make_causal_mask_opt(
|
|
|
|
|
|
|
|
|
293 |
if self.bidirectional_mask is not None:
|
294 |
assert attention_mask.shape == self.bidirectional_mask.shape
|
295 |
-
expanded_bidirectional_mask = _expand_mask_opt(
|
296 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
297 |
if attention_mask is not None:
|
298 |
-
expanded_attn_mask = _expand_mask_opt(
|
299 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
300 |
return combined_attention_mask
|
301 |
-
setattr(model.model.decoder, '_prepare_decoder_attention_mask', MethodType(_prepare_decoder_attention_mask, model.model.decoder))
|
302 |
-
|
303 |
-
def forward(self: OPTForCausalLM, input_ids: Optional[torch.LongTensor]=None, attention_mask: Optional[torch.Tensor]=None, bidirectional_mask: Optional[torch.ByteTensor]=None, head_mask: Optional[torch.Tensor]=None, past_key_values: Optional[List[torch.FloatTensor]]=None, inputs_embeds: Optional[torch.FloatTensor]=None, labels: Optional[torch.LongTensor]=None, use_cache: Optional[bool]=None, output_attentions: Optional[bool]=None, output_hidden_states: Optional[bool]=None, return_dict: Optional[bool]=None):
|
304 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
305 |
def call_og_forward():
|
306 |
-
return self._original_forward(
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
307 |
if bidirectional_mask is None:
|
308 |
return call_og_forward()
|
309 |
self.model.decoder.bidirectional_mask = bidirectional_mask
|
@@ -317,7 +635,7 @@ def _convert_opt_causal_lm_to_prefix_lm(model: OPTForCausalLM) -> OPTForCausalLM
|
|
317 |
|
318 |
def generate(self: OPTForCausalLM, *args: tuple, **kwargs: Dict[str, Any]):
|
319 |
"""Wraps original generate to enable PrefixLM-style attention."""
|
320 |
-
self.model.decoder.bidirectional_mask =
|
321 |
try:
|
322 |
output = self._original_generate(*args, **kwargs)
|
323 |
except:
|
@@ -325,12 +643,23 @@ def _convert_opt_causal_lm_to_prefix_lm(model: OPTForCausalLM) -> OPTForCausalLM
|
|
325 |
raise
|
326 |
self.model.decoder.bidirectional_mask = None
|
327 |
return output
|
328 |
-
|
329 |
-
setattr(model,
|
330 |
-
setattr(model,
|
|
|
331 |
return model
|
|
|
|
|
332 |
_SUPPORTED_HF_MODELS = _SUPPORTED_GPT_MODELS + (BloomForCausalLM, OPTForCausalLM)
|
333 |
-
CAUSAL_LM_TYPES = Union[
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
334 |
|
335 |
def convert_hf_causal_lm_to_prefix_lm(model: CAUSAL_LM_TYPES) -> CAUSAL_LM_TYPES:
|
336 |
"""Converts a HuggingFace Causal LM to a Prefix LM.
|
@@ -396,7 +725,12 @@ def convert_hf_causal_lm_to_prefix_lm(model: CAUSAL_LM_TYPES) -> CAUSAL_LM_TYPES
|
|
396 |
elif isinstance(model, OPTForCausalLM):
|
397 |
return _convert_opt_causal_lm_to_prefix_lm(model)
|
398 |
else:
|
399 |
-
raise TypeError(
|
|
|
|
|
|
|
|
|
|
|
400 |
|
401 |
def add_bidirectional_mask_if_missing(batch: Dict[str, Any]):
|
402 |
"""Attempts to add bidirectional_mask to batch if missing.
|
@@ -404,12 +738,16 @@ def add_bidirectional_mask_if_missing(batch: Dict[str, Any]):
|
|
404 |
Raises:
|
405 |
KeyError if bidirectional_mask is missing and can't be inferred
|
406 |
"""
|
407 |
-
if
|
408 |
-
if batch.get(
|
409 |
-
batch[
|
410 |
-
for
|
411 |
-
batch[
|
412 |
-
elif
|
413 |
-
batch[
|
|
|
|
|
414 |
else:
|
415 |
-
raise KeyError(
|
|
|
|
|
|
11 |
from types import MethodType
|
12 |
from typing import Any, Dict, List, Optional, Tuple, Union
|
13 |
import torch
|
14 |
+
from transformers.models.bloom.modeling_bloom import (
|
15 |
+
BaseModelOutputWithPastAndCrossAttentions,
|
16 |
+
BloomForCausalLM,
|
17 |
+
BloomModel,
|
18 |
+
CausalLMOutputWithCrossAttentions,
|
19 |
+
CrossEntropyLoss,
|
20 |
+
)
|
21 |
from transformers.models.bloom.modeling_bloom import _expand_mask as _expand_mask_bloom
|
22 |
+
from transformers.models.bloom.modeling_bloom import (
|
23 |
+
_make_causal_mask as _make_causal_mask_bloom,
|
24 |
+
)
|
25 |
from transformers.models.bloom.modeling_bloom import logging
|
26 |
from transformers.models.gpt2.modeling_gpt2 import GPT2LMHeadModel
|
27 |
from transformers.models.gpt_neo.modeling_gpt_neo import GPTNeoForCausalLM
|
|
|
29 |
from transformers.models.gptj.modeling_gptj import GPTJForCausalLM
|
30 |
from transformers.models.opt.modeling_opt import OPTForCausalLM
|
31 |
from transformers.models.opt.modeling_opt import _expand_mask as _expand_mask_opt
|
32 |
+
from transformers.models.opt.modeling_opt import (
|
33 |
+
_make_causal_mask as _make_causal_mask_opt,
|
34 |
+
)
|
35 |
+
|
36 |
logger = logging.get_logger(__name__)
|
37 |
+
_SUPPORTED_GPT_MODELS = (
|
38 |
+
GPT2LMHeadModel,
|
39 |
+
GPTJForCausalLM,
|
40 |
+
GPTNeoForCausalLM,
|
41 |
+
GPTNeoXForCausalLM,
|
42 |
+
)
|
43 |
+
CAUSAL_GPT_TYPES = Union[
|
44 |
+
GPT2LMHeadModel, GPTJForCausalLM, GPTNeoForCausalLM, GPTNeoXForCausalLM
|
45 |
+
]
|
46 |
+
|
47 |
|
48 |
def _convert_gpt_causal_lm_to_prefix_lm(model: CAUSAL_GPT_TYPES) -> CAUSAL_GPT_TYPES:
|
49 |
"""Converts a GPT-style Causal LM to a Prefix LM.
|
|
|
56 |
|
57 |
See `convert_hf_causal_lm_to_prefix_lm` for more details.
|
58 |
"""
|
59 |
+
if hasattr(model, "_prefix_lm_converted"):
|
60 |
return model
|
61 |
assert isinstance(model, _SUPPORTED_GPT_MODELS)
|
62 |
+
assert (
|
63 |
+
model.config.add_cross_attention == False
|
64 |
+
), "Only supports GPT-style decoder-only models"
|
65 |
|
66 |
def _get_attn_modules(model: CAUSAL_GPT_TYPES) -> List[torch.nn.Module]:
|
67 |
"""Helper that gets a list of the model's attention modules.
|
|
|
77 |
blocks = model.transformer.h
|
78 |
for block in blocks:
|
79 |
if isinstance(model, GPTNeoForCausalLM):
|
80 |
+
if block.attn.attention_type != "global":
|
81 |
continue
|
82 |
attn_module = block.attn.attention
|
83 |
elif isinstance(model, GPTNeoXForCausalLM):
|
|
|
86 |
attn_module = block.attn
|
87 |
attn_modules.append(attn_module)
|
88 |
return attn_modules
|
|
|
|
|
89 |
|
90 |
+
setattr(model, "_original_forward", getattr(model, "forward"))
|
91 |
+
setattr(model, "_original_generate", getattr(model, "generate"))
|
92 |
+
|
93 |
+
def forward(
|
94 |
+
self: CAUSAL_GPT_TYPES,
|
95 |
+
input_ids: Optional[torch.LongTensor] = None,
|
96 |
+
past_key_values: Optional[Tuple[Tuple[torch.Tensor]]] = None,
|
97 |
+
attention_mask: Optional[torch.FloatTensor] = None,
|
98 |
+
bidirectional_mask: Optional[torch.Tensor] = None,
|
99 |
+
token_type_ids: Optional[torch.LongTensor] = None,
|
100 |
+
position_ids: Optional[torch.LongTensor] = None,
|
101 |
+
head_mask: Optional[torch.FloatTensor] = None,
|
102 |
+
inputs_embeds: Optional[torch.FloatTensor] = None,
|
103 |
+
labels: Optional[torch.LongTensor] = None,
|
104 |
+
use_cache: Optional[bool] = None,
|
105 |
+
output_attentions: Optional[bool] = None,
|
106 |
+
output_hidden_states: Optional[bool] = None,
|
107 |
+
return_dict: Optional[bool] = None,
|
108 |
+
):
|
109 |
"""Wraps original forward to enable PrefixLM attention."""
|
110 |
|
111 |
def call_og_forward():
|
112 |
if isinstance(self, GPTNeoXForCausalLM):
|
113 |
+
return self._original_forward(
|
114 |
+
input_ids=input_ids,
|
115 |
+
past_key_values=past_key_values,
|
116 |
+
attention_mask=attention_mask,
|
117 |
+
head_mask=head_mask,
|
118 |
+
inputs_embeds=inputs_embeds,
|
119 |
+
labels=labels,
|
120 |
+
use_cache=use_cache,
|
121 |
+
output_attentions=output_attentions,
|
122 |
+
output_hidden_states=output_hidden_states,
|
123 |
+
return_dict=return_dict,
|
124 |
+
)
|
125 |
else:
|
126 |
+
return self._original_forward(
|
127 |
+
input_ids=input_ids,
|
128 |
+
past_key_values=past_key_values,
|
129 |
+
attention_mask=attention_mask,
|
130 |
+
token_type_ids=token_type_ids,
|
131 |
+
position_ids=position_ids,
|
132 |
+
head_mask=head_mask,
|
133 |
+
inputs_embeds=inputs_embeds,
|
134 |
+
labels=labels,
|
135 |
+
use_cache=use_cache,
|
136 |
+
output_attentions=output_attentions,
|
137 |
+
output_hidden_states=output_hidden_states,
|
138 |
+
return_dict=return_dict,
|
139 |
+
)
|
140 |
+
|
141 |
if bidirectional_mask is None:
|
142 |
return call_og_forward()
|
143 |
assert isinstance(bidirectional_mask, torch.Tensor)
|
|
|
145 |
(b, s) = bidirectional_mask.shape
|
146 |
max_length = attn_modules[0].bias.shape[-1]
|
147 |
if s > max_length:
|
148 |
+
raise ValueError(
|
149 |
+
f"bidirectional_mask sequence length (={s}) exceeds the "
|
150 |
+
+ f"max length allowed by the model ({max_length})."
|
151 |
+
)
|
152 |
assert s <= max_length
|
153 |
if s < max_length:
|
154 |
+
pad = torch.zeros(
|
155 |
+
(int(b), int(max_length - s)),
|
156 |
+
dtype=bidirectional_mask.dtype,
|
157 |
+
device=bidirectional_mask.device,
|
158 |
+
)
|
159 |
bidirectional_mask = torch.cat([bidirectional_mask, pad], dim=1)
|
160 |
bidirectional = bidirectional_mask.unsqueeze(1).unsqueeze(1)
|
161 |
for attn_module in attn_modules:
|
162 |
+
attn_module.bias.data = torch.logical_or(
|
163 |
+
attn_module.bias.data, bidirectional
|
164 |
+
)
|
165 |
output = call_og_forward()
|
166 |
for attn_module in attn_modules:
|
167 |
attn_module.bias.data = torch.tril(attn_module.bias.data[0, 0])[None, None]
|
|
|
176 |
for attn_module in attn_modules:
|
177 |
attn_module.bias.data = torch.tril(attn_module.bias.data[0, 0])[None, None]
|
178 |
return output
|
179 |
+
|
180 |
+
setattr(model, "forward", MethodType(forward, model))
|
181 |
+
setattr(model, "generate", MethodType(generate, model))
|
182 |
+
setattr(model, "_prefix_lm_converted", True)
|
183 |
return model
|
184 |
|
185 |
+
|
186 |
def _convert_bloom_causal_lm_to_prefix_lm(model: BloomForCausalLM) -> BloomForCausalLM:
|
187 |
"""Converts a BLOOM Causal LM to a Prefix LM.
|
188 |
|
|
|
191 |
|
192 |
See `convert_hf_causal_lm_to_prefix_lm` for more details.
|
193 |
"""
|
194 |
+
if hasattr(model, "_prefix_lm_converted"):
|
195 |
return model
|
196 |
assert isinstance(model, BloomForCausalLM)
|
197 |
+
assert (
|
198 |
+
model.config.add_cross_attention == False
|
199 |
+
), "Only supports BLOOM decoder-only models"
|
200 |
+
|
201 |
+
def _prepare_attn_mask(
|
202 |
+
self: BloomModel,
|
203 |
+
attention_mask: torch.Tensor,
|
204 |
+
bidirectional_mask: Optional[torch.Tensor],
|
205 |
+
input_shape: Tuple[int, int],
|
206 |
+
past_key_values_length: int,
|
207 |
+
) -> torch.BoolTensor:
|
208 |
combined_attention_mask = None
|
209 |
device = attention_mask.device
|
210 |
(_, src_length) = input_shape
|
211 |
if src_length > 1:
|
212 |
+
combined_attention_mask = _make_causal_mask_bloom(
|
213 |
+
input_shape,
|
214 |
+
device=device,
|
215 |
+
past_key_values_length=past_key_values_length,
|
216 |
+
)
|
217 |
if bidirectional_mask is not None:
|
218 |
assert attention_mask.shape == bidirectional_mask.shape
|
219 |
+
expanded_bidirectional_mask = _expand_mask_bloom(
|
220 |
+
bidirectional_mask, tgt_length=src_length
|
221 |
+
)
|
222 |
+
combined_attention_mask = torch.logical_and(
|
223 |
+
combined_attention_mask, expanded_bidirectional_mask
|
224 |
+
)
|
225 |
expanded_attn_mask = _expand_mask_bloom(attention_mask, tgt_length=src_length)
|
226 |
+
combined_attention_mask = (
|
227 |
+
expanded_attn_mask
|
228 |
+
if combined_attention_mask is None
|
229 |
+
else expanded_attn_mask | combined_attention_mask
|
230 |
+
)
|
231 |
return combined_attention_mask
|
232 |
|
233 |
+
def _build_alibi_tensor(
|
234 |
+
self: BloomModel,
|
235 |
+
batch_size: int,
|
236 |
+
query_length: int,
|
237 |
+
key_length: int,
|
238 |
+
dtype: torch.dtype,
|
239 |
+
device: torch.device,
|
240 |
+
) -> torch.Tensor:
|
241 |
num_heads = self.config.n_head
|
242 |
closest_power_of_2 = 2 ** math.floor(math.log2(num_heads))
|
243 |
+
base = torch.tensor(
|
244 |
+
2 ** (-(2 ** (-(math.log2(closest_power_of_2) - 3)))),
|
245 |
+
device=device,
|
246 |
+
dtype=torch.float32,
|
247 |
+
)
|
248 |
+
powers = torch.arange(
|
249 |
+
1, 1 + closest_power_of_2, device=device, dtype=torch.int32
|
250 |
+
)
|
251 |
slopes = torch.pow(base, powers)
|
252 |
if closest_power_of_2 != num_heads:
|
253 |
+
extra_base = torch.tensor(
|
254 |
+
2 ** (-(2 ** (-(math.log2(2 * closest_power_of_2) - 3)))),
|
255 |
+
device=device,
|
256 |
+
dtype=torch.float32,
|
257 |
+
)
|
258 |
+
num_remaining_heads = min(
|
259 |
+
closest_power_of_2, num_heads - closest_power_of_2
|
260 |
+
)
|
261 |
+
extra_powers = torch.arange(
|
262 |
+
1, 1 + 2 * num_remaining_heads, 2, device=device, dtype=torch.int32
|
263 |
+
)
|
264 |
slopes = torch.cat([slopes, torch.pow(extra_base, extra_powers)], dim=0)
|
265 |
qa = torch.arange(query_length, device=device, dtype=torch.int32).view(-1, 1)
|
266 |
ka = torch.arange(key_length, device=device, dtype=torch.int32).view(1, -1)
|
267 |
diffs = qa - ka + key_length - query_length
|
268 |
diffs = -diffs.abs()
|
269 |
+
alibi = slopes.view(1, num_heads, 1, 1) * diffs.view(
|
270 |
+
1, 1, query_length, key_length
|
271 |
+
)
|
272 |
+
alibi = alibi.expand(batch_size, -1, -1, -1).reshape(
|
273 |
+
-1, query_length, key_length
|
274 |
+
)
|
275 |
return alibi.to(dtype)
|
276 |
+
|
277 |
KeyValueT = Tuple[torch.Tensor, torch.Tensor]
|
278 |
|
279 |
+
def forward(
|
280 |
+
self: BloomModel,
|
281 |
+
input_ids: Optional[torch.LongTensor] = None,
|
282 |
+
past_key_values: Optional[Tuple[KeyValueT, ...]] = None,
|
283 |
+
attention_mask: Optional[torch.Tensor] = None,
|
284 |
+
bidirectional_mask: Optional[torch.Tensor] = None,
|
285 |
+
head_mask: Optional[torch.LongTensor] = None,
|
286 |
+
inputs_embeds: Optional[torch.LongTensor] = None,
|
287 |
+
use_cache: Optional[bool] = None,
|
288 |
+
output_attentions: Optional[bool] = None,
|
289 |
+
output_hidden_states: Optional[bool] = None,
|
290 |
+
return_dict: Optional[bool] = None,
|
291 |
+
**deprecated_arguments,
|
292 |
+
) -> Union[Tuple[torch.Tensor, ...], BaseModelOutputWithPastAndCrossAttentions]:
|
293 |
+
if deprecated_arguments.pop("position_ids", False) is not False:
|
294 |
+
warnings.warn(
|
295 |
+
"`position_ids` have no functionality in BLOOM and will be removed in v5.0.0. "
|
296 |
+
+ "You can safely ignore passing `position_ids`.",
|
297 |
+
FutureWarning,
|
298 |
+
)
|
299 |
if len(deprecated_arguments) > 0:
|
300 |
+
raise ValueError(f"Got unexpected arguments: {deprecated_arguments}")
|
301 |
+
output_attentions = (
|
302 |
+
output_attentions
|
303 |
+
if output_attentions is not None
|
304 |
+
else self.config.output_attentions
|
305 |
+
)
|
306 |
+
output_hidden_states = (
|
307 |
+
output_hidden_states
|
308 |
+
if output_hidden_states is not None
|
309 |
+
else self.config.output_hidden_states
|
310 |
+
)
|
311 |
use_cache = use_cache if use_cache is not None else self.config.use_cache
|
312 |
+
return_dict = (
|
313 |
+
return_dict if return_dict is not None else self.config.use_return_dict
|
314 |
+
)
|
315 |
if input_ids is not None and inputs_embeds is not None:
|
316 |
+
raise ValueError(
|
317 |
+
"You cannot specify both input_ids and inputs_embeds at the same time"
|
318 |
+
)
|
319 |
elif input_ids is not None:
|
320 |
(batch_size, seq_length) = input_ids.shape
|
321 |
elif inputs_embeds is not None:
|
322 |
(batch_size, seq_length, _) = inputs_embeds.shape
|
323 |
else:
|
324 |
+
raise ValueError("You have to specify either input_ids or inputs_embeds")
|
325 |
if past_key_values is None:
|
326 |
past_key_values = tuple([None] * len(self.h))
|
327 |
head_mask = self.get_head_mask(head_mask, self.config.n_layer)
|
|
|
338 |
past_key_values_length = tmp.shape[2]
|
339 |
seq_length_with_past = seq_length_with_past + past_key_values_length
|
340 |
if attention_mask is None:
|
341 |
+
attention_mask = torch.ones(
|
342 |
+
(batch_size, seq_length_with_past), device=hidden_states.device
|
343 |
+
)
|
344 |
else:
|
345 |
attention_mask = attention_mask.to(hidden_states.device)
|
346 |
+
alibi = self._build_alibi_tensor(
|
347 |
+
batch_size=batch_size,
|
348 |
+
query_length=seq_length,
|
349 |
+
key_length=seq_length_with_past,
|
350 |
+
dtype=hidden_states.dtype,
|
351 |
+
device=hidden_states.device,
|
352 |
+
)
|
353 |
+
causal_mask = self._prepare_attn_mask(
|
354 |
+
attention_mask,
|
355 |
+
bidirectional_mask,
|
356 |
+
input_shape=(batch_size, seq_length),
|
357 |
+
past_key_values_length=past_key_values_length,
|
358 |
+
)
|
359 |
+
for i, (block, layer_past) in enumerate(zip(self.h, past_key_values)):
|
360 |
if output_hidden_states:
|
361 |
hst = (hidden_states,)
|
362 |
all_hidden_states = all_hidden_states + hst
|
363 |
if self.gradient_checkpointing and self.training:
|
364 |
if use_cache:
|
365 |
+
logger.warning(
|
366 |
+
"`use_cache=True` is incompatible with gradient checkpointing. Setting `use_cache=False`..."
|
367 |
+
)
|
368 |
use_cache = False
|
369 |
|
370 |
def create_custom_forward(module):
|
|
|
371 |
def custom_forward(*inputs):
|
372 |
+
return module(
|
373 |
+
*inputs,
|
374 |
+
use_cache=use_cache,
|
375 |
+
output_attentions=output_attentions,
|
376 |
+
)
|
377 |
+
|
378 |
return custom_forward
|
379 |
+
|
380 |
+
outputs = torch.utils.checkpoint.checkpoint(
|
381 |
+
create_custom_forward(block),
|
382 |
+
hidden_states,
|
383 |
+
alibi,
|
384 |
+
causal_mask,
|
385 |
+
head_mask[i],
|
386 |
+
)
|
387 |
else:
|
388 |
+
outputs = block(
|
389 |
+
hidden_states,
|
390 |
+
layer_past=layer_past,
|
391 |
+
attention_mask=causal_mask,
|
392 |
+
head_mask=head_mask[i],
|
393 |
+
use_cache=use_cache,
|
394 |
+
output_attentions=output_attentions,
|
395 |
+
alibi=alibi,
|
396 |
+
)
|
397 |
hidden_states = outputs[0]
|
398 |
if use_cache is True:
|
399 |
presents = presents + (outputs[1],)
|
|
|
405 |
hst = (hidden_states,)
|
406 |
all_hidden_states = all_hidden_states + hst
|
407 |
if not return_dict:
|
408 |
+
return tuple(
|
409 |
+
(
|
410 |
+
v
|
411 |
+
for v in [
|
412 |
+
hidden_states,
|
413 |
+
presents,
|
414 |
+
all_hidden_states,
|
415 |
+
all_self_attentions,
|
416 |
+
]
|
417 |
+
if v is not None
|
418 |
+
)
|
419 |
+
)
|
420 |
+
return BaseModelOutputWithPastAndCrossAttentions(
|
421 |
+
last_hidden_state=hidden_states,
|
422 |
+
past_key_values=presents,
|
423 |
+
hidden_states=all_hidden_states,
|
424 |
+
attentions=all_self_attentions,
|
425 |
+
)
|
426 |
+
|
427 |
+
setattr(
|
428 |
+
model.transformer,
|
429 |
+
"_prepare_attn_mask",
|
430 |
+
MethodType(_prepare_attn_mask, model.transformer),
|
431 |
+
)
|
432 |
+
setattr(
|
433 |
+
model.transformer,
|
434 |
+
"_build_alibi_tensor",
|
435 |
+
MethodType(_build_alibi_tensor, model.transformer),
|
436 |
+
)
|
437 |
+
setattr(model.transformer, "forward", MethodType(forward, model.transformer))
|
438 |
KeyValueT = Tuple[torch.Tensor, torch.Tensor]
|
439 |
|
440 |
+
def forward(
|
441 |
+
self: BloomForCausalLM,
|
442 |
+
input_ids: Optional[torch.LongTensor] = None,
|
443 |
+
past_key_values: Optional[Tuple[KeyValueT, ...]] = None,
|
444 |
+
attention_mask: Optional[torch.Tensor] = None,
|
445 |
+
bidirectional_mask: Optional[torch.Tensor] = None,
|
446 |
+
head_mask: Optional[torch.Tensor] = None,
|
447 |
+
inputs_embeds: Optional[torch.Tensor] = None,
|
448 |
+
labels: Optional[torch.Tensor] = None,
|
449 |
+
use_cache: Optional[bool] = None,
|
450 |
+
output_attentions: Optional[bool] = None,
|
451 |
+
output_hidden_states: Optional[bool] = None,
|
452 |
+
return_dict: Optional[bool] = None,
|
453 |
+
**deprecated_arguments,
|
454 |
+
) -> Union[Tuple[torch.Tensor], CausalLMOutputWithCrossAttentions]:
|
455 |
"""Replacement forward method for BloomCausalLM."""
|
456 |
+
if deprecated_arguments.pop("position_ids", False) is not False:
|
457 |
+
warnings.warn(
|
458 |
+
"`position_ids` have no functionality in BLOOM and will be removed "
|
459 |
+
+ "in v5.0.0. You can safely ignore passing `position_ids`.",
|
460 |
+
FutureWarning,
|
461 |
+
)
|
462 |
if len(deprecated_arguments) > 0:
|
463 |
+
raise ValueError(f"Got unexpected arguments: {deprecated_arguments}")
|
464 |
+
return_dict = (
|
465 |
+
return_dict if return_dict is not None else self.config.use_return_dict
|
466 |
+
)
|
467 |
+
transformer_outputs = self.transformer(
|
468 |
+
input_ids,
|
469 |
+
past_key_values=past_key_values,
|
470 |
+
attention_mask=attention_mask,
|
471 |
+
bidirectional_mask=bidirectional_mask,
|
472 |
+
head_mask=head_mask,
|
473 |
+
inputs_embeds=inputs_embeds,
|
474 |
+
use_cache=use_cache,
|
475 |
+
output_attentions=output_attentions,
|
476 |
+
output_hidden_states=output_hidden_states,
|
477 |
+
return_dict=return_dict,
|
478 |
+
)
|
479 |
hidden_states = transformer_outputs[0]
|
480 |
lm_logits = self.lm_head(hidden_states)
|
481 |
loss = None
|
|
|
484 |
shift_labels = labels[..., 1:].contiguous()
|
485 |
(batch_size, seq_length, vocab_size) = shift_logits.shape
|
486 |
loss_fct = CrossEntropyLoss()
|
487 |
+
loss = loss_fct(
|
488 |
+
shift_logits.view(batch_size * seq_length, vocab_size),
|
489 |
+
shift_labels.view(batch_size * seq_length),
|
490 |
+
)
|
491 |
if not return_dict:
|
492 |
output = (lm_logits,) + transformer_outputs[1:]
|
493 |
return (loss,) + output if loss is not None else output
|
494 |
+
return CausalLMOutputWithCrossAttentions(
|
495 |
+
loss=loss,
|
496 |
+
logits=lm_logits,
|
497 |
+
past_key_values=transformer_outputs.past_key_values,
|
498 |
+
hidden_states=transformer_outputs.hidden_states,
|
499 |
+
attentions=transformer_outputs.attentions,
|
500 |
+
)
|
501 |
+
|
502 |
+
def prepare_inputs_for_generation(
|
503 |
+
self: BloomForCausalLM,
|
504 |
+
input_ids: torch.LongTensor,
|
505 |
+
past: Optional[torch.Tensor] = None,
|
506 |
+
attention_mask: Optional[torch.Tensor] = None,
|
507 |
+
**kwargs,
|
508 |
+
) -> dict:
|
509 |
if past:
|
510 |
input_ids = input_ids[:, -1].unsqueeze(-1)
|
511 |
bidirectional_mask = None
|
|
|
513 |
past = self._convert_to_bloom_cache(past)
|
514 |
else:
|
515 |
bidirectional_mask = torch.ones_like(input_ids)
|
516 |
+
return {
|
517 |
+
"input_ids": input_ids,
|
518 |
+
"past_key_values": past,
|
519 |
+
"use_cache": True,
|
520 |
+
"attention_mask": attention_mask,
|
521 |
+
"bidirectional_mask": bidirectional_mask,
|
522 |
+
}
|
523 |
+
|
524 |
+
setattr(model, "forward", MethodType(forward, model))
|
525 |
+
setattr(
|
526 |
+
model,
|
527 |
+
"prepare_inputs_for_generation",
|
528 |
+
MethodType(prepare_inputs_for_generation, model),
|
529 |
+
)
|
530 |
+
setattr(model, "_prefix_lm_converted", True)
|
531 |
return model
|
532 |
|
533 |
+
|
534 |
def _convert_opt_causal_lm_to_prefix_lm(model: OPTForCausalLM) -> OPTForCausalLM:
|
535 |
"""Converts an OPT Causal LM to a Prefix LM.
|
536 |
|
|
|
539 |
|
540 |
See `convert_hf_causal_lm_to_prefix_lm` for more details.
|
541 |
"""
|
542 |
+
if hasattr(model, "_prefix_lm_converted"):
|
543 |
return model
|
544 |
assert isinstance(model, OPTForCausalLM)
|
545 |
+
assert (
|
546 |
+
model.config.add_cross_attention == False
|
547 |
+
), "Only supports OPT decoder-only models"
|
548 |
+
setattr(model, "_original_forward", getattr(model, "forward"))
|
549 |
+
setattr(model, "_original_generate", getattr(model, "generate"))
|
550 |
model.model.decoder.bidirectional_mask = None
|
551 |
|
552 |
+
def _prepare_decoder_attention_mask(
|
553 |
+
self, attention_mask, input_shape, inputs_embeds, past_key_values_length
|
554 |
+
):
|
555 |
combined_attention_mask = None
|
556 |
if input_shape[-1] > 1:
|
557 |
+
if self.bidirectional_mask == "g":
|
558 |
(bsz, src_length) = input_shape
|
559 |
+
combined_attention_mask = torch.zeros(
|
560 |
+
(bsz, 1, src_length, src_length + past_key_values_length),
|
561 |
+
dtype=inputs_embeds.dtype,
|
562 |
+
device=inputs_embeds.device,
|
563 |
+
)
|
564 |
else:
|
565 |
+
combined_attention_mask = _make_causal_mask_opt(
|
566 |
+
input_shape,
|
567 |
+
inputs_embeds.dtype,
|
568 |
+
past_key_values_length=past_key_values_length,
|
569 |
+
).to(inputs_embeds.device)
|
570 |
if self.bidirectional_mask is not None:
|
571 |
assert attention_mask.shape == self.bidirectional_mask.shape
|
572 |
+
expanded_bidirectional_mask = _expand_mask_opt(
|
573 |
+
self.bidirectional_mask,
|
574 |
+
inputs_embeds.dtype,
|
575 |
+
tgt_len=input_shape[-1],
|
576 |
+
).to(inputs_embeds.device)
|
577 |
+
combined_attention_mask = torch.maximum(
|
578 |
+
expanded_bidirectional_mask, combined_attention_mask
|
579 |
+
)
|
580 |
if attention_mask is not None:
|
581 |
+
expanded_attn_mask = _expand_mask_opt(
|
582 |
+
attention_mask, inputs_embeds.dtype, tgt_len=input_shape[-1]
|
583 |
+
).to(inputs_embeds.device)
|
584 |
+
combined_attention_mask = (
|
585 |
+
expanded_attn_mask
|
586 |
+
if combined_attention_mask is None
|
587 |
+
else expanded_attn_mask + combined_attention_mask
|
588 |
+
)
|
589 |
return combined_attention_mask
|
|
|
|
|
|
|
590 |
|
591 |
+
setattr(
|
592 |
+
model.model.decoder,
|
593 |
+
"_prepare_decoder_attention_mask",
|
594 |
+
MethodType(_prepare_decoder_attention_mask, model.model.decoder),
|
595 |
+
)
|
596 |
+
|
597 |
+
def forward(
|
598 |
+
self: OPTForCausalLM,
|
599 |
+
input_ids: Optional[torch.LongTensor] = None,
|
600 |
+
attention_mask: Optional[torch.Tensor] = None,
|
601 |
+
bidirectional_mask: Optional[torch.ByteTensor] = None,
|
602 |
+
head_mask: Optional[torch.Tensor] = None,
|
603 |
+
past_key_values: Optional[List[torch.FloatTensor]] = None,
|
604 |
+
inputs_embeds: Optional[torch.FloatTensor] = None,
|
605 |
+
labels: Optional[torch.LongTensor] = None,
|
606 |
+
use_cache: Optional[bool] = None,
|
607 |
+
output_attentions: Optional[bool] = None,
|
608 |
+
output_hidden_states: Optional[bool] = None,
|
609 |
+
return_dict: Optional[bool] = None,
|
610 |
+
):
|
611 |
def call_og_forward():
|
612 |
+
return self._original_forward(
|
613 |
+
input_ids=input_ids,
|
614 |
+
attention_mask=attention_mask,
|
615 |
+
head_mask=head_mask,
|
616 |
+
past_key_values=past_key_values,
|
617 |
+
inputs_embeds=inputs_embeds,
|
618 |
+
labels=labels,
|
619 |
+
use_cache=use_cache,
|
620 |
+
output_attentions=output_attentions,
|
621 |
+
output_hidden_states=output_hidden_states,
|
622 |
+
return_dict=return_dict,
|
623 |
+
)
|
624 |
+
|
625 |
if bidirectional_mask is None:
|
626 |
return call_og_forward()
|
627 |
self.model.decoder.bidirectional_mask = bidirectional_mask
|
|
|
635 |
|
636 |
def generate(self: OPTForCausalLM, *args: tuple, **kwargs: Dict[str, Any]):
|
637 |
"""Wraps original generate to enable PrefixLM-style attention."""
|
638 |
+
self.model.decoder.bidirectional_mask = "g"
|
639 |
try:
|
640 |
output = self._original_generate(*args, **kwargs)
|
641 |
except:
|
|
|
643 |
raise
|
644 |
self.model.decoder.bidirectional_mask = None
|
645 |
return output
|
646 |
+
|
647 |
+
setattr(model, "forward", MethodType(forward, model))
|
648 |
+
setattr(model, "generate", MethodType(generate, model))
|
649 |
+
setattr(model, "_prefix_lm_converted", True)
|
650 |
return model
|
651 |
+
|
652 |
+
|
653 |
_SUPPORTED_HF_MODELS = _SUPPORTED_GPT_MODELS + (BloomForCausalLM, OPTForCausalLM)
|
654 |
+
CAUSAL_LM_TYPES = Union[
|
655 |
+
GPT2LMHeadModel,
|
656 |
+
GPTJForCausalLM,
|
657 |
+
GPTNeoForCausalLM,
|
658 |
+
GPTNeoXForCausalLM,
|
659 |
+
BloomForCausalLM,
|
660 |
+
OPTForCausalLM,
|
661 |
+
]
|
662 |
+
|
663 |
|
664 |
def convert_hf_causal_lm_to_prefix_lm(model: CAUSAL_LM_TYPES) -> CAUSAL_LM_TYPES:
|
665 |
"""Converts a HuggingFace Causal LM to a Prefix LM.
|
|
|
725 |
elif isinstance(model, OPTForCausalLM):
|
726 |
return _convert_opt_causal_lm_to_prefix_lm(model)
|
727 |
else:
|
728 |
+
raise TypeError(
|
729 |
+
f"Cannot convert model to Prefix LM. "
|
730 |
+
+ f"Model does not belong to set of supported HF models:"
|
731 |
+
+ f"\n{_SUPPORTED_HF_MODELS}"
|
732 |
+
)
|
733 |
+
|
734 |
|
735 |
def add_bidirectional_mask_if_missing(batch: Dict[str, Any]):
|
736 |
"""Attempts to add bidirectional_mask to batch if missing.
|
|
|
738 |
Raises:
|
739 |
KeyError if bidirectional_mask is missing and can't be inferred
|
740 |
"""
|
741 |
+
if "bidirectional_mask" not in batch:
|
742 |
+
if batch.get("mode", None) == "icl_task":
|
743 |
+
batch["bidirectional_mask"] = batch["attention_mask"].clone()
|
744 |
+
for i, continuation_indices in enumerate(batch["continuation_indices"]):
|
745 |
+
batch["bidirectional_mask"][i, continuation_indices] = 0
|
746 |
+
elif "labels" in batch and "attention_mask" in batch:
|
747 |
+
batch["bidirectional_mask"] = torch.logical_and(
|
748 |
+
torch.eq(batch["attention_mask"], 1), torch.eq(batch["labels"], -100)
|
749 |
+
).type_as(batch["attention_mask"])
|
750 |
else:
|
751 |
+
raise KeyError(
|
752 |
+
"No bidirectional_mask in batch and not sure how to construct one."
|
753 |
+
)
|
meta_init_context.py
CHANGED
@@ -2,8 +2,9 @@ from contextlib import contextmanager
|
|
2 |
import torch
|
3 |
import torch.nn as nn
|
4 |
|
|
|
5 |
@contextmanager
|
6 |
-
def init_empty_weights(include_buffers: bool=False):
|
7 |
"""Meta initialization context manager.
|
8 |
|
9 |
A context manager under which models are initialized with all parameters
|
@@ -30,11 +31,12 @@ def init_empty_weights(include_buffers: bool=False):
|
|
30 |
|
31 |
</Tip>
|
32 |
"""
|
33 |
-
with init_on_device(torch.device(
|
34 |
yield f
|
35 |
|
|
|
36 |
@contextmanager
|
37 |
-
def init_on_device(device: torch.device, include_buffers: bool=False):
|
38 |
"""Device initialization context manager.
|
39 |
|
40 |
A context manager under which models are initialized with all parameters
|
@@ -62,33 +64,47 @@ def init_on_device(device: torch.device, include_buffers: bool=False):
|
|
62 |
if param is not None:
|
63 |
param_cls = type(module._parameters[name])
|
64 |
kwargs = module._parameters[name].__dict__
|
65 |
-
module._parameters[name] = param_cls(
|
|
|
|
|
66 |
|
67 |
def register_empty_buffer(module, name, buffer):
|
68 |
old_register_buffer(module, name, buffer)
|
69 |
if buffer is not None:
|
70 |
module._buffers[name] = module._buffers[name].to(device)
|
|
|
71 |
if include_buffers:
|
72 |
-
tensor_constructors_to_patch = {
|
|
|
|
|
|
|
73 |
else:
|
74 |
tensor_constructors_to_patch = {}
|
75 |
|
76 |
def patch_tensor_constructor(fn):
|
77 |
-
|
78 |
def wrapper(*args, **kwargs):
|
79 |
-
kwargs[
|
80 |
return fn(*args, **kwargs)
|
|
|
81 |
return wrapper
|
|
|
82 |
try:
|
83 |
nn.Module.register_parameter = register_empty_parameter
|
84 |
if include_buffers:
|
85 |
nn.Module.register_buffer = register_empty_buffer
|
86 |
for torch_function_name in tensor_constructors_to_patch.keys():
|
87 |
-
setattr(
|
|
|
|
|
|
|
|
|
88 |
yield
|
89 |
finally:
|
90 |
nn.Module.register_parameter = old_register_parameter
|
91 |
if include_buffers:
|
92 |
nn.Module.register_buffer = old_register_buffer
|
93 |
-
for (
|
94 |
-
|
|
|
|
|
|
|
|
2 |
import torch
|
3 |
import torch.nn as nn
|
4 |
|
5 |
+
|
6 |
@contextmanager
|
7 |
+
def init_empty_weights(include_buffers: bool = False):
|
8 |
"""Meta initialization context manager.
|
9 |
|
10 |
A context manager under which models are initialized with all parameters
|
|
|
31 |
|
32 |
</Tip>
|
33 |
"""
|
34 |
+
with init_on_device(torch.device("meta"), include_buffers=include_buffers) as f:
|
35 |
yield f
|
36 |
|
37 |
+
|
38 |
@contextmanager
|
39 |
+
def init_on_device(device: torch.device, include_buffers: bool = False):
|
40 |
"""Device initialization context manager.
|
41 |
|
42 |
A context manager under which models are initialized with all parameters
|
|
|
64 |
if param is not None:
|
65 |
param_cls = type(module._parameters[name])
|
66 |
kwargs = module._parameters[name].__dict__
|
67 |
+
module._parameters[name] = param_cls(
|
68 |
+
module._parameters[name].to(device), **kwargs
|
69 |
+
)
|
70 |
|
71 |
def register_empty_buffer(module, name, buffer):
|
72 |
old_register_buffer(module, name, buffer)
|
73 |
if buffer is not None:
|
74 |
module._buffers[name] = module._buffers[name].to(device)
|
75 |
+
|
76 |
if include_buffers:
|
77 |
+
tensor_constructors_to_patch = {
|
78 |
+
torch_function_name: getattr(torch, torch_function_name)
|
79 |
+
for torch_function_name in ["empty", "zeros", "ones", "full"]
|
80 |
+
}
|
81 |
else:
|
82 |
tensor_constructors_to_patch = {}
|
83 |
|
84 |
def patch_tensor_constructor(fn):
|
|
|
85 |
def wrapper(*args, **kwargs):
|
86 |
+
kwargs["device"] = device
|
87 |
return fn(*args, **kwargs)
|
88 |
+
|
89 |
return wrapper
|
90 |
+
|
91 |
try:
|
92 |
nn.Module.register_parameter = register_empty_parameter
|
93 |
if include_buffers:
|
94 |
nn.Module.register_buffer = register_empty_buffer
|
95 |
for torch_function_name in tensor_constructors_to_patch.keys():
|
96 |
+
setattr(
|
97 |
+
torch,
|
98 |
+
torch_function_name,
|
99 |
+
patch_tensor_constructor(getattr(torch, torch_function_name)),
|
100 |
+
)
|
101 |
yield
|
102 |
finally:
|
103 |
nn.Module.register_parameter = old_register_parameter
|
104 |
if include_buffers:
|
105 |
nn.Module.register_buffer = old_register_buffer
|
106 |
+
for (
|
107 |
+
torch_function_name,
|
108 |
+
old_torch_function,
|
109 |
+
) in tensor_constructors_to_patch.items():
|
110 |
+
setattr(torch, torch_function_name, old_torch_function)
|
norm.py
CHANGED
@@ -1,28 +1,55 @@
|
|
1 |
import torch
|
2 |
|
|
|
3 |
def _cast_if_autocast_enabled(tensor):
|
4 |
if torch.is_autocast_enabled():
|
5 |
-
if tensor.device.type ==
|
6 |
dtype = torch.get_autocast_gpu_dtype()
|
7 |
-
elif tensor.device.type ==
|
8 |
dtype = torch.get_autocast_cpu_dtype()
|
9 |
else:
|
10 |
raise NotImplementedError()
|
11 |
return tensor.to(dtype=dtype)
|
12 |
return tensor
|
13 |
|
14 |
-
class LPLayerNorm(torch.nn.LayerNorm):
|
15 |
|
16 |
-
|
17 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
18 |
|
19 |
def forward(self, x):
|
20 |
module_device = x.device
|
21 |
downcast_x = _cast_if_autocast_enabled(x)
|
22 |
-
downcast_weight =
|
23 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
24 |
with torch.autocast(enabled=False, device_type=module_device.type):
|
25 |
-
return torch.nn.functional.layer_norm(
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
26 |
|
27 |
def rms_norm(x, weight=None, eps=1e-05):
|
28 |
output = x / torch.rsqrt(x.pow(2).mean(-1, keepdim=True) + eps)
|
@@ -30,27 +57,50 @@ def rms_norm(x, weight=None, eps=1e-05):
|
|
30 |
return output * weight
|
31 |
return output
|
32 |
|
33 |
-
class RMSNorm(torch.nn.Module):
|
34 |
|
35 |
-
|
|
|
|
|
|
|
36 |
super().__init__()
|
37 |
self.eps = eps
|
38 |
if weight:
|
39 |
-
self.weight = torch.nn.Parameter(
|
|
|
|
|
40 |
else:
|
41 |
-
self.register_parameter(
|
42 |
|
43 |
def forward(self, x):
|
44 |
return rms_norm(x.float(), self.weight, self.eps).to(dtype=x.dtype)
|
45 |
|
46 |
-
class LPRMSNorm(RMSNorm):
|
47 |
|
48 |
-
|
49 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
50 |
|
51 |
def forward(self, x):
|
52 |
downcast_x = _cast_if_autocast_enabled(x)
|
53 |
-
downcast_weight =
|
|
|
|
|
|
|
|
|
54 |
with torch.autocast(enabled=False, device_type=x.device.type):
|
55 |
return rms_norm(downcast_x, downcast_weight, self.eps).to(dtype=x.dtype)
|
56 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
import torch
|
2 |
|
3 |
+
|
4 |
def _cast_if_autocast_enabled(tensor):
|
5 |
if torch.is_autocast_enabled():
|
6 |
+
if tensor.device.type == "cuda":
|
7 |
dtype = torch.get_autocast_gpu_dtype()
|
8 |
+
elif tensor.device.type == "cpu":
|
9 |
dtype = torch.get_autocast_cpu_dtype()
|
10 |
else:
|
11 |
raise NotImplementedError()
|
12 |
return tensor.to(dtype=dtype)
|
13 |
return tensor
|
14 |
|
|
|
15 |
|
16 |
+
class LPLayerNorm(torch.nn.LayerNorm):
|
17 |
+
def __init__(
|
18 |
+
self,
|
19 |
+
normalized_shape,
|
20 |
+
eps=1e-05,
|
21 |
+
elementwise_affine=True,
|
22 |
+
device=None,
|
23 |
+
dtype=None,
|
24 |
+
):
|
25 |
+
super().__init__(
|
26 |
+
normalized_shape=normalized_shape,
|
27 |
+
eps=eps,
|
28 |
+
elementwise_affine=elementwise_affine,
|
29 |
+
device=device,
|
30 |
+
dtype=dtype,
|
31 |
+
)
|
32 |
|
33 |
def forward(self, x):
|
34 |
module_device = x.device
|
35 |
downcast_x = _cast_if_autocast_enabled(x)
|
36 |
+
downcast_weight = (
|
37 |
+
_cast_if_autocast_enabled(self.weight)
|
38 |
+
if self.weight is not None
|
39 |
+
else self.weight
|
40 |
+
)
|
41 |
+
downcast_bias = (
|
42 |
+
_cast_if_autocast_enabled(self.bias) if self.bias is not None else self.bias
|
43 |
+
)
|
44 |
with torch.autocast(enabled=False, device_type=module_device.type):
|
45 |
+
return torch.nn.functional.layer_norm(
|
46 |
+
downcast_x,
|
47 |
+
self.normalized_shape,
|
48 |
+
downcast_weight,
|
49 |
+
downcast_bias,
|
50 |
+
self.eps,
|
51 |
+
)
|
52 |
+
|
53 |
|
54 |
def rms_norm(x, weight=None, eps=1e-05):
|
55 |
output = x / torch.rsqrt(x.pow(2).mean(-1, keepdim=True) + eps)
|
|
|
57 |
return output * weight
|
58 |
return output
|
59 |
|
|
|
60 |
|
61 |
+
class RMSNorm(torch.nn.Module):
|
62 |
+
def __init__(
|
63 |
+
self, normalized_shape, eps=1e-05, weight=True, dtype=None, device=None
|
64 |
+
):
|
65 |
super().__init__()
|
66 |
self.eps = eps
|
67 |
if weight:
|
68 |
+
self.weight = torch.nn.Parameter(
|
69 |
+
torch.ones(normalized_shape, dtype=dtype, device=device)
|
70 |
+
)
|
71 |
else:
|
72 |
+
self.register_parameter("weight", None)
|
73 |
|
74 |
def forward(self, x):
|
75 |
return rms_norm(x.float(), self.weight, self.eps).to(dtype=x.dtype)
|
76 |
|
|
|
77 |
|
78 |
+
class LPRMSNorm(RMSNorm):
|
79 |
+
def __init__(
|
80 |
+
self, normalized_shape, eps=1e-05, weight=True, dtype=None, device=None
|
81 |
+
):
|
82 |
+
super().__init__(
|
83 |
+
normalized_shape=normalized_shape,
|
84 |
+
eps=eps,
|
85 |
+
weight=weight,
|
86 |
+
dtype=dtype,
|
87 |
+
device=device,
|
88 |
+
)
|
89 |
|
90 |
def forward(self, x):
|
91 |
downcast_x = _cast_if_autocast_enabled(x)
|
92 |
+
downcast_weight = (
|
93 |
+
_cast_if_autocast_enabled(self.weight)
|
94 |
+
if self.weight is not None
|
95 |
+
else self.weight
|
96 |
+
)
|
97 |
with torch.autocast(enabled=False, device_type=x.device.type):
|
98 |
return rms_norm(downcast_x, downcast_weight, self.eps).to(dtype=x.dtype)
|
99 |
+
|
100 |
+
|
101 |
+
NORM_CLASS_REGISTRY = {
|
102 |
+
"layernorm": torch.nn.LayerNorm,
|
103 |
+
"low_precision_layernorm": LPLayerNorm,
|
104 |
+
"rmsnorm": RMSNorm,
|
105 |
+
"low_precision_rmsnorm": LPRMSNorm,
|
106 |
+
}
|
param_init_fns.py
CHANGED
@@ -7,97 +7,133 @@ import torch
|
|
7 |
from torch import nn
|
8 |
from .norm import NORM_CLASS_REGISTRY
|
9 |
|
10 |
-
|
|
|
11 |
del kwargs
|
12 |
if verbose > 1:
|
13 |
warnings.warn(f"Initializing network using module's reset_parameters attribute")
|
14 |
-
if hasattr(module,
|
15 |
module.reset_parameters()
|
16 |
|
|
|
17 |
def fused_init_helper_(module: nn.Module, init_fn_):
|
18 |
-
_fused = getattr(module,
|
19 |
if _fused is None:
|
20 |
-
raise RuntimeError(f
|
21 |
(dim, splits) = _fused
|
22 |
splits = (0, *splits, module.weight.size(dim))
|
23 |
-
for
|
24 |
slice_indices = [slice(None)] * module.weight.ndim
|
25 |
slice_indices[dim] = slice(s, e)
|
26 |
init_fn_(module.weight[slice_indices])
|
27 |
|
28 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
29 |
del kwargs
|
30 |
if verbose > 1:
|
31 |
-
warnings.warn(f
|
32 |
init_div_is_residual = init_div_is_residual
|
33 |
if init_div_is_residual is False:
|
34 |
div_is_residual = 1.0
|
35 |
elif init_div_is_residual is True:
|
36 |
div_is_residual = math.sqrt(2 * n_layers)
|
37 |
-
elif isinstance(init_div_is_residual, float) or isinstance(
|
|
|
|
|
38 |
div_is_residual = init_div_is_residual
|
39 |
elif isinstance(init_div_is_residual, str) and init_div_is_residual.isnumeric():
|
40 |
div_is_residual = float(init_div_is_residual)
|
41 |
else:
|
42 |
div_is_residual = 1.0
|
43 |
-
raise ValueError(
|
|
|
|
|
44 |
if init_div_is_residual is not False:
|
45 |
if verbose > 1:
|
46 |
-
warnings.warn(
|
|
|
|
|
|
|
47 |
if isinstance(module, nn.Linear):
|
48 |
-
if hasattr(module,
|
49 |
fused_init_helper_(module, init_fn_)
|
50 |
else:
|
51 |
init_fn_(module.weight)
|
52 |
if module.bias is not None:
|
53 |
torch.nn.init.zeros_(module.bias)
|
54 |
-
if init_div_is_residual is not False and getattr(module,
|
55 |
with torch.no_grad():
|
56 |
module.weight.div_(div_is_residual)
|
57 |
elif isinstance(module, nn.Embedding):
|
58 |
if emb_init_std is not None:
|
59 |
std = emb_init_std
|
60 |
if std == 0:
|
61 |
-
warnings.warn(f
|
62 |
emb_init_fn_ = partial(torch.nn.init.normal_, mean=0.0, std=std)
|
63 |
if verbose > 1:
|
64 |
-
warnings.warn(
|
|
|
|
|
65 |
elif emb_init_uniform_lim is not None:
|
66 |
lim = emb_init_uniform_lim
|
67 |
if isinstance(lim, Sequence):
|
68 |
if len(lim) > 2:
|
69 |
-
raise ValueError(
|
|
|
|
|
70 |
if lim[0] == lim[1]:
|
71 |
-
warnings.warn(f
|
72 |
else:
|
73 |
if lim == 0:
|
74 |
-
warnings.warn(f
|
75 |
lim = [-lim, lim]
|
76 |
(a, b) = lim
|
77 |
emb_init_fn_ = partial(torch.nn.init.uniform_, a=a, b=b)
|
78 |
if verbose > 1:
|
79 |
-
warnings.warn(
|
|
|
|
|
80 |
else:
|
81 |
emb_init_fn_ = init_fn_
|
82 |
emb_init_fn_(module.weight)
|
83 |
elif isinstance(module, tuple(set(NORM_CLASS_REGISTRY.values()))):
|
84 |
if verbose > 1:
|
85 |
-
warnings.warn(
|
86 |
-
|
|
|
|
|
87 |
torch.nn.init.ones_(module.weight)
|
88 |
-
if hasattr(module,
|
89 |
torch.nn.init.zeros_(module.bias)
|
90 |
elif isinstance(module, nn.MultiheadAttention):
|
91 |
if module._qkv_same_embed_dim:
|
92 |
assert module.in_proj_weight is not None
|
93 |
-
assert
|
|
|
|
|
|
|
|
|
94 |
assert d_model is not None
|
95 |
_d = d_model
|
96 |
splits = (0, _d, 2 * _d, 3 * _d)
|
97 |
-
for
|
98 |
init_fn_(module.in_proj_weight[s:e])
|
99 |
else:
|
100 |
-
assert
|
|
|
|
|
|
|
|
|
101 |
assert module.in_proj_weight is None
|
102 |
init_fn_(module.q_proj_weight)
|
103 |
init_fn_(module.k_proj_weight)
|
@@ -109,37 +145,112 @@ def generic_param_init_fn_(module: nn.Module, init_fn_, n_layers: int, d_model:
|
|
109 |
if module.bias_v is not None:
|
110 |
torch.nn.init.zeros_(module.bias_v)
|
111 |
init_fn_(module.out_proj.weight)
|
112 |
-
if init_div_is_residual is not False and getattr(
|
|
|
|
|
113 |
with torch.no_grad():
|
114 |
module.out_proj.weight.div_(div_is_residual)
|
115 |
if module.out_proj.bias is not None:
|
116 |
torch.nn.init.zeros_(module.out_proj.bias)
|
117 |
else:
|
118 |
for _ in module.parameters(recurse=False):
|
119 |
-
raise NotImplementedError(
|
|
|
|
|
|
|
120 |
|
121 |
def _normal_init_(std, mean=0.0):
|
122 |
return partial(torch.nn.init.normal_, mean=mean, std=std)
|
123 |
|
124 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
125 |
del kwargs
|
126 |
init_fn_ = _normal_init_(std=std)
|
127 |
if verbose > 1:
|
128 |
-
warnings.warn(f
|
129 |
-
generic_param_init_fn_(
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
130 |
|
131 |
-
def baseline_param_init_fn_(
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
132 |
del kwargs
|
133 |
if init_std is None:
|
134 |
-
raise ValueError(
|
135 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
136 |
|
137 |
-
def small_param_init_fn_(
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
138 |
del kwargs
|
139 |
std = math.sqrt(2 / (5 * d_model))
|
140 |
-
_normal_param_init_fn_(
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
141 |
|
142 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
143 |
"""From section 2.3.1 of GPT-NeoX-20B:
|
144 |
|
145 |
An Open-Source AutoregressiveLanguage Model — Black et. al. (2022)
|
@@ -149,33 +260,158 @@ def neox_param_init_fn_(module: nn.Module, n_layers: int, d_model: int, emb_init
|
|
149 |
del kwargs
|
150 |
residual_div = n_layers / math.sqrt(10)
|
151 |
if verbose > 1:
|
152 |
-
warnings.warn(f
|
153 |
-
small_param_init_fn_(
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
154 |
|
155 |
-
def kaiming_uniform_param_init_fn_(
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
156 |
del kwargs
|
157 |
if verbose > 1:
|
158 |
-
warnings.warn(
|
159 |
-
|
160 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
161 |
|
162 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
163 |
del kwargs
|
164 |
if verbose > 1:
|
165 |
-
warnings.warn(
|
166 |
-
|
167 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
168 |
|
169 |
-
def xavier_uniform_param_init_fn_(
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
170 |
del kwargs
|
171 |
xavier_uniform_ = partial(torch.nn.init.xavier_uniform_, gain=init_gain)
|
172 |
if verbose > 1:
|
173 |
-
warnings.warn(
|
174 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
175 |
|
176 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
177 |
xavier_normal_ = partial(torch.nn.init.xavier_normal_, gain=init_gain)
|
178 |
if verbose > 1:
|
179 |
-
warnings.warn(
|
180 |
-
|
181 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
7 |
from torch import nn
|
8 |
from .norm import NORM_CLASS_REGISTRY
|
9 |
|
10 |
+
|
11 |
+
def torch_default_param_init_fn_(module: nn.Module, verbose: int = 0, **kwargs):
|
12 |
del kwargs
|
13 |
if verbose > 1:
|
14 |
warnings.warn(f"Initializing network using module's reset_parameters attribute")
|
15 |
+
if hasattr(module, "reset_parameters"):
|
16 |
module.reset_parameters()
|
17 |
|
18 |
+
|
19 |
def fused_init_helper_(module: nn.Module, init_fn_):
|
20 |
+
_fused = getattr(module, "_fused", None)
|
21 |
if _fused is None:
|
22 |
+
raise RuntimeError(f"Internal logic error")
|
23 |
(dim, splits) = _fused
|
24 |
splits = (0, *splits, module.weight.size(dim))
|
25 |
+
for s, e in zip(splits[:-1], splits[1:]):
|
26 |
slice_indices = [slice(None)] * module.weight.ndim
|
27 |
slice_indices[dim] = slice(s, e)
|
28 |
init_fn_(module.weight[slice_indices])
|
29 |
|
30 |
+
|
31 |
+
def generic_param_init_fn_(
|
32 |
+
module: nn.Module,
|
33 |
+
init_fn_,
|
34 |
+
n_layers: int,
|
35 |
+
d_model: Optional[int] = None,
|
36 |
+
init_div_is_residual: Union[int, float, str, bool] = True,
|
37 |
+
emb_init_std: Optional[float] = None,
|
38 |
+
emb_init_uniform_lim: Optional[Union[Tuple[float, float], float]] = None,
|
39 |
+
verbose: int = 0,
|
40 |
+
**kwargs,
|
41 |
+
):
|
42 |
del kwargs
|
43 |
if verbose > 1:
|
44 |
+
warnings.warn(f"If model has bias parameters they are initialized to 0.")
|
45 |
init_div_is_residual = init_div_is_residual
|
46 |
if init_div_is_residual is False:
|
47 |
div_is_residual = 1.0
|
48 |
elif init_div_is_residual is True:
|
49 |
div_is_residual = math.sqrt(2 * n_layers)
|
50 |
+
elif isinstance(init_div_is_residual, float) or isinstance(
|
51 |
+
init_div_is_residual, int
|
52 |
+
):
|
53 |
div_is_residual = init_div_is_residual
|
54 |
elif isinstance(init_div_is_residual, str) and init_div_is_residual.isnumeric():
|
55 |
div_is_residual = float(init_div_is_residual)
|
56 |
else:
|
57 |
div_is_residual = 1.0
|
58 |
+
raise ValueError(
|
59 |
+
f"Expected init_div_is_residual to be boolean or numeric, got {init_div_is_residual}"
|
60 |
+
)
|
61 |
if init_div_is_residual is not False:
|
62 |
if verbose > 1:
|
63 |
+
warnings.warn(
|
64 |
+
f"Initializing _is_residual layers then dividing them by {div_is_residual:.3f}. "
|
65 |
+
+ f"Set `init_div_is_residual: false` in init config to disable this."
|
66 |
+
)
|
67 |
if isinstance(module, nn.Linear):
|
68 |
+
if hasattr(module, "_fused"):
|
69 |
fused_init_helper_(module, init_fn_)
|
70 |
else:
|
71 |
init_fn_(module.weight)
|
72 |
if module.bias is not None:
|
73 |
torch.nn.init.zeros_(module.bias)
|
74 |
+
if init_div_is_residual is not False and getattr(module, "_is_residual", False):
|
75 |
with torch.no_grad():
|
76 |
module.weight.div_(div_is_residual)
|
77 |
elif isinstance(module, nn.Embedding):
|
78 |
if emb_init_std is not None:
|
79 |
std = emb_init_std
|
80 |
if std == 0:
|
81 |
+
warnings.warn(f"Embedding layer initialized to 0.")
|
82 |
emb_init_fn_ = partial(torch.nn.init.normal_, mean=0.0, std=std)
|
83 |
if verbose > 1:
|
84 |
+
warnings.warn(
|
85 |
+
f"Embedding layer initialized using normal distribution with mean=0 and std={std!r}."
|
86 |
+
)
|
87 |
elif emb_init_uniform_lim is not None:
|
88 |
lim = emb_init_uniform_lim
|
89 |
if isinstance(lim, Sequence):
|
90 |
if len(lim) > 2:
|
91 |
+
raise ValueError(
|
92 |
+
f"Uniform init requires a min and a max limit. User input: {lim}."
|
93 |
+
)
|
94 |
if lim[0] == lim[1]:
|
95 |
+
warnings.warn(f"Embedding layer initialized to {lim[0]}.")
|
96 |
else:
|
97 |
if lim == 0:
|
98 |
+
warnings.warn(f"Embedding layer initialized to 0.")
|
99 |
lim = [-lim, lim]
|
100 |
(a, b) = lim
|
101 |
emb_init_fn_ = partial(torch.nn.init.uniform_, a=a, b=b)
|
102 |
if verbose > 1:
|
103 |
+
warnings.warn(
|
104 |
+
f"Embedding layer initialized using uniform distribution in range {lim}."
|
105 |
+
)
|
106 |
else:
|
107 |
emb_init_fn_ = init_fn_
|
108 |
emb_init_fn_(module.weight)
|
109 |
elif isinstance(module, tuple(set(NORM_CLASS_REGISTRY.values()))):
|
110 |
if verbose > 1:
|
111 |
+
warnings.warn(
|
112 |
+
f"Norm weights are set to 1. If norm layer has a bias it is initialized to 0."
|
113 |
+
)
|
114 |
+
if hasattr(module, "weight") and module.weight is not None:
|
115 |
torch.nn.init.ones_(module.weight)
|
116 |
+
if hasattr(module, "bias") and module.bias is not None:
|
117 |
torch.nn.init.zeros_(module.bias)
|
118 |
elif isinstance(module, nn.MultiheadAttention):
|
119 |
if module._qkv_same_embed_dim:
|
120 |
assert module.in_proj_weight is not None
|
121 |
+
assert (
|
122 |
+
module.q_proj_weight is None
|
123 |
+
and module.k_proj_weight is None
|
124 |
+
and (module.v_proj_weight is None)
|
125 |
+
)
|
126 |
assert d_model is not None
|
127 |
_d = d_model
|
128 |
splits = (0, _d, 2 * _d, 3 * _d)
|
129 |
+
for s, e in zip(splits[:-1], splits[1:]):
|
130 |
init_fn_(module.in_proj_weight[s:e])
|
131 |
else:
|
132 |
+
assert (
|
133 |
+
module.q_proj_weight is not None
|
134 |
+
and module.k_proj_weight is not None
|
135 |
+
and (module.v_proj_weight is not None)
|
136 |
+
)
|
137 |
assert module.in_proj_weight is None
|
138 |
init_fn_(module.q_proj_weight)
|
139 |
init_fn_(module.k_proj_weight)
|
|
|
145 |
if module.bias_v is not None:
|
146 |
torch.nn.init.zeros_(module.bias_v)
|
147 |
init_fn_(module.out_proj.weight)
|
148 |
+
if init_div_is_residual is not False and getattr(
|
149 |
+
module.out_proj, "_is_residual", False
|
150 |
+
):
|
151 |
with torch.no_grad():
|
152 |
module.out_proj.weight.div_(div_is_residual)
|
153 |
if module.out_proj.bias is not None:
|
154 |
torch.nn.init.zeros_(module.out_proj.bias)
|
155 |
else:
|
156 |
for _ in module.parameters(recurse=False):
|
157 |
+
raise NotImplementedError(
|
158 |
+
f"{module.__class__.__name__} parameters are not initialized by param_init_fn."
|
159 |
+
)
|
160 |
+
|
161 |
|
162 |
def _normal_init_(std, mean=0.0):
|
163 |
return partial(torch.nn.init.normal_, mean=mean, std=std)
|
164 |
|
165 |
+
|
166 |
+
def _normal_param_init_fn_(
|
167 |
+
module: nn.Module,
|
168 |
+
std: float,
|
169 |
+
n_layers: int,
|
170 |
+
d_model: Optional[int] = None,
|
171 |
+
init_div_is_residual: Union[int, float, str, bool] = True,
|
172 |
+
emb_init_std: Optional[float] = None,
|
173 |
+
emb_init_uniform_lim: Optional[Union[Tuple[float, float], float]] = None,
|
174 |
+
verbose: int = 0,
|
175 |
+
**kwargs,
|
176 |
+
):
|
177 |
del kwargs
|
178 |
init_fn_ = _normal_init_(std=std)
|
179 |
if verbose > 1:
|
180 |
+
warnings.warn(f"Using torch.nn.init.normal_ init fn mean=0.0, std={std}")
|
181 |
+
generic_param_init_fn_(
|
182 |
+
module=module,
|
183 |
+
init_fn_=init_fn_,
|
184 |
+
d_model=d_model,
|
185 |
+
n_layers=n_layers,
|
186 |
+
init_div_is_residual=init_div_is_residual,
|
187 |
+
emb_init_std=emb_init_std,
|
188 |
+
emb_init_uniform_lim=emb_init_uniform_lim,
|
189 |
+
verbose=verbose,
|
190 |
+
)
|
191 |
+
|
192 |
|
193 |
+
def baseline_param_init_fn_(
|
194 |
+
module: nn.Module,
|
195 |
+
init_std: float,
|
196 |
+
n_layers: int,
|
197 |
+
d_model: Optional[int] = None,
|
198 |
+
init_div_is_residual: Union[int, float, str, bool] = True,
|
199 |
+
emb_init_std: Optional[float] = None,
|
200 |
+
emb_init_uniform_lim: Optional[Union[Tuple[float, float], float]] = None,
|
201 |
+
verbose: int = 0,
|
202 |
+
**kwargs,
|
203 |
+
):
|
204 |
del kwargs
|
205 |
if init_std is None:
|
206 |
+
raise ValueError(
|
207 |
+
"You must set model.init_config['init_std'] to a float value to use the default initialization scheme."
|
208 |
+
)
|
209 |
+
_normal_param_init_fn_(
|
210 |
+
module=module,
|
211 |
+
std=init_std,
|
212 |
+
d_model=d_model,
|
213 |
+
n_layers=n_layers,
|
214 |
+
init_div_is_residual=init_div_is_residual,
|
215 |
+
emb_init_std=emb_init_std,
|
216 |
+
emb_init_uniform_lim=emb_init_uniform_lim,
|
217 |
+
verbose=verbose,
|
218 |
+
)
|
219 |
+
|
220 |
|
221 |
+
def small_param_init_fn_(
|
222 |
+
module: nn.Module,
|
223 |
+
n_layers: int,
|
224 |
+
d_model: int,
|
225 |
+
init_div_is_residual: Union[int, float, str, bool] = True,
|
226 |
+
emb_init_std: Optional[float] = None,
|
227 |
+
emb_init_uniform_lim: Optional[Union[Tuple[float, float], float]] = None,
|
228 |
+
verbose: int = 0,
|
229 |
+
**kwargs,
|
230 |
+
):
|
231 |
del kwargs
|
232 |
std = math.sqrt(2 / (5 * d_model))
|
233 |
+
_normal_param_init_fn_(
|
234 |
+
module=module,
|
235 |
+
std=std,
|
236 |
+
d_model=d_model,
|
237 |
+
n_layers=n_layers,
|
238 |
+
init_div_is_residual=init_div_is_residual,
|
239 |
+
emb_init_std=emb_init_std,
|
240 |
+
emb_init_uniform_lim=emb_init_uniform_lim,
|
241 |
+
verbose=verbose,
|
242 |
+
)
|
243 |
|
244 |
+
|
245 |
+
def neox_param_init_fn_(
|
246 |
+
module: nn.Module,
|
247 |
+
n_layers: int,
|
248 |
+
d_model: int,
|
249 |
+
emb_init_std: Optional[float] = None,
|
250 |
+
emb_init_uniform_lim: Optional[Union[Tuple[float, float], float]] = None,
|
251 |
+
verbose: int = 0,
|
252 |
+
**kwargs,
|
253 |
+
):
|
254 |
"""From section 2.3.1 of GPT-NeoX-20B:
|
255 |
|
256 |
An Open-Source AutoregressiveLanguage Model — Black et. al. (2022)
|
|
|
260 |
del kwargs
|
261 |
residual_div = n_layers / math.sqrt(10)
|
262 |
if verbose > 1:
|
263 |
+
warnings.warn(f"setting init_div_is_residual to {residual_div}")
|
264 |
+
small_param_init_fn_(
|
265 |
+
module=module,
|
266 |
+
d_model=d_model,
|
267 |
+
n_layers=n_layers,
|
268 |
+
init_div_is_residual=residual_div,
|
269 |
+
emb_init_std=emb_init_std,
|
270 |
+
emb_init_uniform_lim=emb_init_uniform_lim,
|
271 |
+
verbose=verbose,
|
272 |
+
)
|
273 |
+
|
274 |
|
275 |
+
def kaiming_uniform_param_init_fn_(
|
276 |
+
module: nn.Module,
|
277 |
+
n_layers: int,
|
278 |
+
d_model: Optional[int] = None,
|
279 |
+
init_div_is_residual: Union[int, float, str, bool] = True,
|
280 |
+
emb_init_std: Optional[float] = None,
|
281 |
+
emb_init_uniform_lim: Optional[Union[Tuple[float, float], float]] = None,
|
282 |
+
init_gain: float = 0,
|
283 |
+
fan_mode: str = "fan_in",
|
284 |
+
init_nonlinearity: str = "leaky_relu",
|
285 |
+
verbose: int = 0,
|
286 |
+
**kwargs,
|
287 |
+
):
|
288 |
del kwargs
|
289 |
if verbose > 1:
|
290 |
+
warnings.warn(
|
291 |
+
f"Using nn.init.kaiming_uniform_ init fn with parameters: "
|
292 |
+
+ f"a={init_gain}, mode={fan_mode}, nonlinearity={init_nonlinearity}"
|
293 |
+
)
|
294 |
+
kaiming_uniform_ = partial(
|
295 |
+
nn.init.kaiming_uniform_,
|
296 |
+
a=init_gain,
|
297 |
+
mode=fan_mode,
|
298 |
+
nonlinearity=init_nonlinearity,
|
299 |
+
)
|
300 |
+
generic_param_init_fn_(
|
301 |
+
module=module,
|
302 |
+
init_fn_=kaiming_uniform_,
|
303 |
+
d_model=d_model,
|
304 |
+
n_layers=n_layers,
|
305 |
+
init_div_is_residual=init_div_is_residual,
|
306 |
+
emb_init_std=emb_init_std,
|
307 |
+
emb_init_uniform_lim=emb_init_uniform_lim,
|
308 |
+
verbose=verbose,
|
309 |
+
)
|
310 |
|
311 |
+
|
312 |
+
def kaiming_normal_param_init_fn_(
|
313 |
+
module: nn.Module,
|
314 |
+
n_layers: int,
|
315 |
+
d_model: Optional[int] = None,
|
316 |
+
init_div_is_residual: Union[int, float, str, bool] = True,
|
317 |
+
emb_init_std: Optional[float] = None,
|
318 |
+
emb_init_uniform_lim: Optional[Union[Tuple[float, float], float]] = None,
|
319 |
+
init_gain: float = 0,
|
320 |
+
fan_mode: str = "fan_in",
|
321 |
+
init_nonlinearity: str = "leaky_relu",
|
322 |
+
verbose: int = 0,
|
323 |
+
**kwargs,
|
324 |
+
):
|
325 |
del kwargs
|
326 |
if verbose > 1:
|
327 |
+
warnings.warn(
|
328 |
+
f"Using nn.init.kaiming_normal_ init fn with parameters: "
|
329 |
+
+ f"a={init_gain}, mode={fan_mode}, nonlinearity={init_nonlinearity}"
|
330 |
+
)
|
331 |
+
kaiming_normal_ = partial(
|
332 |
+
torch.nn.init.kaiming_normal_,
|
333 |
+
a=init_gain,
|
334 |
+
mode=fan_mode,
|
335 |
+
nonlinearity=init_nonlinearity,
|
336 |
+
)
|
337 |
+
generic_param_init_fn_(
|
338 |
+
module=module,
|
339 |
+
init_fn_=kaiming_normal_,
|
340 |
+
d_model=d_model,
|
341 |
+
n_layers=n_layers,
|
342 |
+
init_div_is_residual=init_div_is_residual,
|
343 |
+
emb_init_std=emb_init_std,
|
344 |
+
emb_init_uniform_lim=emb_init_uniform_lim,
|
345 |
+
verbose=verbose,
|
346 |
+
)
|
347 |
+
|
348 |
|
349 |
+
def xavier_uniform_param_init_fn_(
|
350 |
+
module: nn.Module,
|
351 |
+
n_layers: int,
|
352 |
+
d_model: Optional[int] = None,
|
353 |
+
init_div_is_residual: Union[int, float, str, bool] = True,
|
354 |
+
emb_init_std: Optional[float] = None,
|
355 |
+
emb_init_uniform_lim: Optional[Union[Tuple[float, float], float]] = None,
|
356 |
+
init_gain: float = 0,
|
357 |
+
verbose: int = 0,
|
358 |
+
**kwargs,
|
359 |
+
):
|
360 |
del kwargs
|
361 |
xavier_uniform_ = partial(torch.nn.init.xavier_uniform_, gain=init_gain)
|
362 |
if verbose > 1:
|
363 |
+
warnings.warn(
|
364 |
+
f"Using torch.nn.init.xavier_uniform_ init fn with parameters: "
|
365 |
+
+ f"gain={init_gain}"
|
366 |
+
)
|
367 |
+
generic_param_init_fn_(
|
368 |
+
module=module,
|
369 |
+
init_fn_=xavier_uniform_,
|
370 |
+
d_model=d_model,
|
371 |
+
n_layers=n_layers,
|
372 |
+
init_div_is_residual=init_div_is_residual,
|
373 |
+
emb_init_std=emb_init_std,
|
374 |
+
emb_init_uniform_lim=emb_init_uniform_lim,
|
375 |
+
verbose=verbose,
|
376 |
+
)
|
377 |
|
378 |
+
|
379 |
+
def xavier_normal_param_init_fn_(
|
380 |
+
module: nn.Module,
|
381 |
+
n_layers: int,
|
382 |
+
d_model: Optional[int] = None,
|
383 |
+
init_div_is_residual: Union[int, float, str, bool] = True,
|
384 |
+
emb_init_std: Optional[float] = None,
|
385 |
+
emb_init_uniform_lim: Optional[Union[Tuple[float, float], float]] = None,
|
386 |
+
init_gain: float = 0,
|
387 |
+
verbose: int = 0,
|
388 |
+
**kwargs,
|
389 |
+
):
|
390 |
xavier_normal_ = partial(torch.nn.init.xavier_normal_, gain=init_gain)
|
391 |
if verbose > 1:
|
392 |
+
warnings.warn(
|
393 |
+
f"Using torch.nn.init.xavier_normal_ init fn with parameters: "
|
394 |
+
+ f"gain={init_gain}"
|
395 |
+
)
|
396 |
+
generic_param_init_fn_(
|
397 |
+
module=module,
|
398 |
+
init_fn_=xavier_normal_,
|
399 |
+
d_model=d_model,
|
400 |
+
n_layers=n_layers,
|
401 |
+
init_div_is_residual=init_div_is_residual,
|
402 |
+
emb_init_std=emb_init_std,
|
403 |
+
emb_init_uniform_lim=emb_init_uniform_lim,
|
404 |
+
verbose=verbose,
|
405 |
+
)
|
406 |
+
|
407 |
+
|
408 |
+
MODEL_INIT_REGISTRY = {
|
409 |
+
"default_": torch_default_param_init_fn_,
|
410 |
+
"baseline_": baseline_param_init_fn_,
|
411 |
+
"kaiming_uniform_": kaiming_uniform_param_init_fn_,
|
412 |
+
"kaiming_normal_": kaiming_normal_param_init_fn_,
|
413 |
+
"neox_init_": neox_param_init_fn_,
|
414 |
+
"small_init_": small_param_init_fn_,
|
415 |
+
"xavier_uniform_": xavier_uniform_param_init_fn_,
|
416 |
+
"xavier_normal_": xavier_normal_param_init_fn_,
|
417 |
+
}
|