OpenNLPLab
commited on
Commit
•
460b22e
1
Parent(s):
cf95141
Fix issues regarding to transformer version
Browse files- generation_config.json +5 -2
- modeling_transnormer.py +145 -279
- tokenization_baichuan.py +5 -5
generation_config.json
CHANGED
@@ -1,6 +1,9 @@
|
|
1 |
{
|
2 |
-
"
|
3 |
"bos_token_id": 1,
|
4 |
"eos_token_id": 2,
|
5 |
-
"
|
|
|
|
|
|
|
6 |
}
|
|
|
1 |
{
|
2 |
+
"pad_token_id": 0,
|
3 |
"bos_token_id": 1,
|
4 |
"eos_token_id": 2,
|
5 |
+
"max_new_tokens": 2048,
|
6 |
+
"temperature": 1.0,
|
7 |
+
"repetition_penalty": 1.03,
|
8 |
+
"do_sample": true
|
9 |
}
|
modeling_transnormer.py
CHANGED
@@ -11,8 +11,7 @@
|
|
11 |
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
12 |
# See the License for the specific language governing permissions and
|
13 |
# limitations under the License.
|
14 |
-
|
15 |
-
# coding=utf-8
|
16 |
""" PyTorch Transnormer model."""
|
17 |
import math
|
18 |
import os
|
@@ -29,7 +28,6 @@ from transformers.activations import ACT2FN
|
|
29 |
from transformers.modeling_outputs import (
|
30 |
BaseModelOutputWithPast,
|
31 |
CausalLMOutputWithPast,
|
32 |
-
SequenceClassifierOutputWithPast,
|
33 |
)
|
34 |
from transformers.modeling_utils import PreTrainedModel
|
35 |
from transformers.utils import (
|
@@ -85,7 +83,6 @@ if not has_lightning_attention:
|
|
85 |
########## start Transnormer
|
86 |
##### Linearized Relative Positional Encoding: https://openreview.net/forum?id=xoLyps2qWc&referrer=%5BAuthor%20Console%5D(%2Fgroup%3Fid%3DTMLR%2FAuthors%23your-submissions)
|
87 |
class Lrpe(nn.Module):
|
88 |
-
|
89 |
def __init__(
|
90 |
self,
|
91 |
num_heads=8,
|
@@ -95,8 +92,9 @@ class Lrpe(nn.Module):
|
|
95 |
d = num_heads * embed_dim
|
96 |
|
97 |
self.index = torch.empty(0)
|
98 |
-
self.theta = nn.Parameter(
|
99 |
-
num_heads, 1, -1)
|
|
|
100 |
|
101 |
def extra_repr(self):
|
102 |
return print_module(self)
|
@@ -115,7 +113,6 @@ class Lrpe(nn.Module):
|
|
115 |
|
116 |
|
117 |
class GLU(nn.Module):
|
118 |
-
|
119 |
def __init__(self, d1, d2, bias=False):
|
120 |
super().__init__()
|
121 |
if debug:
|
@@ -138,7 +135,6 @@ class GLU(nn.Module):
|
|
138 |
|
139 |
|
140 |
class NormLinearAttention(nn.Module):
|
141 |
-
|
142 |
def __init__(
|
143 |
self,
|
144 |
embed_dim,
|
@@ -194,7 +190,7 @@ class NormLinearAttention(nn.Module):
|
|
194 |
output_attentions,
|
195 |
past_key_value,
|
196 |
use_cache,
|
197 |
-
slope_rate
|
198 |
)
|
199 |
# x: b n d
|
200 |
n = x.shape[-2]
|
@@ -202,8 +198,8 @@ class NormLinearAttention(nn.Module):
|
|
202 |
q, k, v, u = self.qkvu_proj(x).chunk(4, dim=-1)
|
203 |
# reshape
|
204 |
q, k, v = map(
|
205 |
-
lambda x: rearrange(x, "b n (h d) -> b h n d", h=self.num_heads),
|
206 |
-
|
207 |
# act
|
208 |
q = self.act(q)
|
209 |
k = self.act(k)
|
@@ -211,7 +207,7 @@ class NormLinearAttention(nn.Module):
|
|
211 |
q_offset = 0
|
212 |
# lrpe relys on position, get cache first
|
213 |
if past_key_value is not None:
|
214 |
-
# reuse k, v,
|
215 |
k = torch.cat([past_key_value[0], k], dim=-2)
|
216 |
v = torch.cat([past_key_value[1], v], dim=-2)
|
217 |
q_offset = past_key_value[0].shape[-2]
|
@@ -228,17 +224,17 @@ class NormLinearAttention(nn.Module):
|
|
228 |
|
229 |
if attn_padding_mask is not None:
|
230 |
v = v.masked_fill(
|
231 |
-
(1 - attn_padding_mask).unsqueeze(1).unsqueeze(-1).to(
|
232 |
-
|
233 |
|
234 |
if not has_lightning_attention:
|
235 |
if slope_rate != None:
|
236 |
attn_mask = torch.exp(slope_rate * attn_mask)
|
237 |
-
|
238 |
output = linear_attention(q, k, v, attn_mask)
|
239 |
else:
|
240 |
-
output = lightning_attention(
|
241 |
-
|
|
|
242 |
|
243 |
# reshape
|
244 |
output = rearrange(output, "b h n d -> b n (h d)")
|
@@ -257,14 +253,14 @@ class NormLinearAttention(nn.Module):
|
|
257 |
return output, attn_weights, past_key_value
|
258 |
|
259 |
def inference(
|
260 |
-
|
261 |
-
|
262 |
-
|
263 |
-
|
264 |
-
|
265 |
-
|
266 |
-
|
267 |
-
|
268 |
):
|
269 |
# x: b n d
|
270 |
n = x.shape[-2]
|
@@ -272,8 +268,8 @@ class NormLinearAttention(nn.Module):
|
|
272 |
q, k, v, u = self.qkvu_proj(x).chunk(4, dim=-1)
|
273 |
# reshape
|
274 |
q, k, v = map(
|
275 |
-
lambda x: rearrange(x, "b n (h d) -> b h n d", h=self.num_heads),
|
276 |
-
|
277 |
# act
|
278 |
q = self.act(q)
|
279 |
k = self.act(k)
|
@@ -281,7 +277,7 @@ class NormLinearAttention(nn.Module):
|
|
281 |
# rpe
|
282 |
if self.linear_use_lrpe:
|
283 |
q = self.lrpe(q, offset=self.offset)
|
284 |
-
k = self.lrpe(k
|
285 |
|
286 |
if past_key_value == None:
|
287 |
self.offset = q.shape[-2]
|
@@ -299,8 +295,7 @@ class NormLinearAttention(nn.Module):
|
|
299 |
|
300 |
if attn_padding_mask is not None:
|
301 |
attn_mask = attn_mask.masked_fill(
|
302 |
-
(1 - attn_padding_mask).unsqueeze(1).unsqueeze(2).to(
|
303 |
-
torch.bool),
|
304 |
0,
|
305 |
)
|
306 |
energy = torch.einsum("... n d, ... m d -> ... n m", q, k)
|
@@ -311,18 +306,17 @@ class NormLinearAttention(nn.Module):
|
|
311 |
output = torch.einsum("... n m, ... m d -> ... n d", energy, v)
|
312 |
|
313 |
eval_and_not_generate = eval(
|
314 |
-
os.environ.get("eval_and_not_generate", default="False")
|
|
|
315 |
if eval_and_not_generate:
|
316 |
kv = None
|
317 |
else:
|
318 |
# b, h, n, e, d
|
319 |
-
kv_outproduct = torch.einsum("... n e, ... n d -> ... n e d",
|
320 |
-
k, v)
|
321 |
# 1, 1, n, 1, 1
|
322 |
-
index = torch.arange(n - 1, -1, -1).reshape(1, 1, -1, 1,
|
323 |
-
1).to(x)
|
324 |
# (h, 1, 1) -> (1, h, 1, 1, 1); (1, h, 1, 1, 1), (1, 1, n, 1, 1) -> (1, h, n, 1, 1)
|
325 |
-
decay = ratio.unsqueeze(0).unsqueeze(-1)**index
|
326 |
|
327 |
kv_outproduct_with_decay = kv_outproduct * decay
|
328 |
kv = torch.sum(kv_outproduct_with_decay, dim=-3)
|
@@ -333,11 +327,12 @@ class NormLinearAttention(nn.Module):
|
|
333 |
for i in range(n):
|
334 |
kv = ratio * kv + torch.einsum(
|
335 |
"... n d, ... n e -> ... d e",
|
336 |
-
k[:, :, i:i + 1],
|
337 |
-
v[:, :, i:i + 1],
|
|
|
|
|
|
|
338 |
)
|
339 |
-
qkv = torch.einsum("... n e, ... e d -> ... n d",
|
340 |
-
q[:, :, i:i + 1], kv)
|
341 |
output.append(qkv)
|
342 |
output = torch.concat(output, dim=-2)
|
343 |
|
@@ -356,7 +351,6 @@ class NormLinearAttention(nn.Module):
|
|
356 |
|
357 |
|
358 |
class TransnormerDecoderLayer(nn.Module):
|
359 |
-
|
360 |
def __init__(self, config: TransnormerConfig):
|
361 |
super().__init__()
|
362 |
self.embed_dim = config.decoder_embed_dim
|
@@ -395,14 +389,14 @@ class TransnormerDecoderLayer(nn.Module):
|
|
395 |
return residual + x
|
396 |
|
397 |
def forward(
|
398 |
-
|
399 |
-
|
400 |
-
|
401 |
-
|
402 |
-
|
403 |
-
|
404 |
-
|
405 |
-
|
406 |
):
|
407 |
residual = x
|
408 |
x = self.token_norm(x)
|
@@ -422,13 +416,13 @@ class TransnormerDecoderLayer(nn.Module):
|
|
422 |
x = self.channel_mixer(x)
|
423 |
x = self.residual_connection(x, residual)
|
424 |
|
425 |
-
outputs = (x,
|
426 |
|
427 |
if output_attentions:
|
428 |
-
outputs += (self_attn_weights,
|
429 |
|
430 |
if use_cache:
|
431 |
-
outputs += (present_key_value,
|
432 |
|
433 |
return outputs
|
434 |
|
@@ -450,7 +444,9 @@ TRANSNORMER_START_DOCSTRING = r"""
|
|
450 |
"""
|
451 |
|
452 |
|
453 |
-
@add_start_docstrings(
|
|
|
|
|
454 |
class TransnormerPreTrainedModel(PreTrainedModel):
|
455 |
config_class = TransnormerConfig
|
456 |
base_model_prefix = "model"
|
@@ -535,7 +531,9 @@ TRANSNORMER_INPUTS_DOCSTRING = r"""
|
|
535 |
"""
|
536 |
|
537 |
|
538 |
-
@add_start_docstrings(
|
|
|
|
|
539 |
class TransnormerModel(TransnormerPreTrainedModel):
|
540 |
"""
|
541 |
Transformer decoder consisting of *config.num_hidden_layers* layers. Each layer is a [`TransnormerDecoderLayer`]
|
@@ -559,31 +557,29 @@ class TransnormerModel(TransnormerPreTrainedModel):
|
|
559 |
self.slopes = self._build_slope_tensor(config.decoder_attention_heads)
|
560 |
|
561 |
# params
|
562 |
-
self.embed_tokens = nn.Embedding(
|
563 |
-
|
564 |
-
|
565 |
self.layers = nn.ModuleList([])
|
566 |
for i in range(config.decoder_layers):
|
567 |
if len(self.linear_use_lrpe_list) > 0:
|
568 |
config.linear_use_lrpe = self.linear_use_lrpe_list[i]
|
569 |
self.layers.append(TransnormerDecoderLayer(config))
|
570 |
|
571 |
-
self.final_norm = get_norm_fn(config.norm_type)(
|
572 |
-
config.decoder_embed_dim)
|
573 |
self.embed_dim = config.decoder_embed_dim
|
574 |
-
self.embed_scale = (
|
575 |
-
self.embed_dim)
|
|
|
576 |
|
577 |
# Initialize weights and apply final processing
|
578 |
self.post_init()
|
579 |
|
580 |
@staticmethod
|
581 |
def _build_slope_tensor(n_attention_heads: int):
|
582 |
-
|
583 |
def get_slopes(n):
|
584 |
-
|
585 |
def get_slopes_power_of_2(n):
|
586 |
-
start = 2**(-(2
|
587 |
ratio = start
|
588 |
return [start * ratio**i for i in range(n)]
|
589 |
|
@@ -592,15 +588,18 @@ class TransnormerModel(TransnormerPreTrainedModel):
|
|
592 |
n
|
593 |
) # In the paper, we only train models that have 2^a heads for some a. This function has
|
594 |
else: # some good properties that only occur when the input is a power of 2. To maintain that even
|
595 |
-
closest_power_of_2 = 2**math.floor(
|
596 |
math.log2(n)
|
597 |
) # when the number of heads is not a power of 2, we use this workaround.
|
598 |
-
return (
|
599 |
-
|
|
|
|
|
600 |
|
601 |
# h, 1, 1
|
602 |
slopes = torch.tensor(get_slopes(n_attention_heads)).reshape(
|
603 |
-
n_attention_heads, 1, 1
|
|
|
604 |
|
605 |
return slopes
|
606 |
|
@@ -613,26 +612,26 @@ class TransnormerModel(TransnormerPreTrainedModel):
|
|
613 |
def set_input_embeddings(self, value):
|
614 |
self.embed_tokens = value
|
615 |
|
616 |
-
def _prepare_decoder_linear_attn_mask(
|
617 |
-
|
|
|
618 |
bsz, tgt_len = input_shape
|
619 |
src_len = tgt_len + past_key_values_length
|
620 |
|
621 |
def power_log(x):
|
622 |
-
return 2**(math.ceil(math.log(x, 2)))
|
623 |
|
624 |
n = power_log(max(tgt_len, src_len))
|
625 |
if self._linear_attn_mask.shape[-1] < n:
|
626 |
|
627 |
def get_mask(n):
|
628 |
-
mask = torch.triu(
|
629 |
-
torch.zeros(n, n).float().fill_(float("-inf")), 1)
|
630 |
# no slope version
|
631 |
# -n, ..., -2, -1, 0
|
632 |
for i in range(n):
|
633 |
x = torch.arange(i + 1)
|
634 |
y = x
|
635 |
-
mask[i, :i + 1] = -torch.flip(y, [0])
|
636 |
|
637 |
return mask
|
638 |
|
@@ -644,8 +643,7 @@ class TransnormerModel(TransnormerPreTrainedModel):
|
|
644 |
linear_attn_mask = self._linear_attn_mask[:, -tgt_len:, -src_len:]
|
645 |
num_heads = linear_attn_mask.shape[0]
|
646 |
|
647 |
-
return linear_attn_mask[None, :, :, :].expand(bsz, num_heads, tgt_len,
|
648 |
-
src_len)
|
649 |
|
650 |
@add_start_docstrings_to_model_forward(TRANSNORMER_INPUTS_DOCSTRING)
|
651 |
def forward(
|
@@ -659,15 +657,21 @@ class TransnormerModel(TransnormerPreTrainedModel):
|
|
659 |
output_hidden_states: Optional[bool] = None,
|
660 |
return_dict: Optional[bool] = None,
|
661 |
) -> Union[Tuple, BaseModelOutputWithPast]:
|
662 |
-
output_attentions = (
|
663 |
-
|
664 |
-
|
665 |
-
|
666 |
-
|
|
|
|
|
|
|
|
|
|
|
667 |
use_cache = use_cache if use_cache is not None else self.config.use_cache
|
668 |
|
669 |
-
return_dict = (
|
670 |
-
|
|
|
671 |
|
672 |
# retrieve input_ids and inputs_embeds
|
673 |
if input_ids is not None and inputs_embeds is not None:
|
@@ -689,7 +693,7 @@ class TransnormerModel(TransnormerPreTrainedModel):
|
|
689 |
if past_key_values is not None:
|
690 |
past_key_values_length = past_key_values[0][0].shape[-2]
|
691 |
seq_length_with_past = seq_length_with_past + past_key_values_length
|
692 |
-
|
693 |
if inputs_embeds is None:
|
694 |
# !!! use embed_scale
|
695 |
inputs_embeds = self.embed_scale * self.embed_tokens(input_ids)
|
@@ -711,72 +715,54 @@ class TransnormerModel(TransnormerPreTrainedModel):
|
|
711 |
##### norm linear layers
|
712 |
linear_attn_padding_mask = attn_padding_mask
|
713 |
linear_attn_mask = self._prepare_decoder_linear_attn_mask(
|
714 |
-
(batch_size, seq_length), inputs_embeds, past_key_values_length
|
|
|
715 |
|
716 |
-
slope_rates = [
|
717 |
-
self.slopes.to(input_ids.device) for _ in range(self.num_layers)
|
718 |
-
]
|
719 |
|
720 |
for idx, layer in enumerate(self.layers):
|
721 |
if output_hidden_states:
|
722 |
-
all_hidden_states += (hidden_states,
|
723 |
|
724 |
-
past_key_value = (
|
725 |
-
|
|
|
726 |
|
727 |
slope_rate = slope_rates[idx]
|
728 |
slope_rate = slope_rate * (1 - idx / (self.num_layers - 1) + 1e-5)
|
729 |
mask = linear_attn_mask
|
730 |
-
|
731 |
-
|
732 |
-
|
733 |
-
|
734 |
-
|
735 |
-
|
736 |
-
|
737 |
-
|
738 |
-
|
739 |
-
|
740 |
-
|
741 |
-
layer_outputs = torch.utils.checkpoint.checkpoint(
|
742 |
-
create_custom_forward(layer),
|
743 |
-
hidden_states,
|
744 |
-
mask,
|
745 |
-
linear_attn_padding_mask,
|
746 |
-
None,
|
747 |
-
)
|
748 |
-
else:
|
749 |
-
layer_outputs = layer(
|
750 |
-
hidden_states,
|
751 |
-
attn_mask=mask,
|
752 |
-
attn_padding_mask=linear_attn_padding_mask,
|
753 |
-
past_key_value=past_key_value,
|
754 |
-
output_attentions=output_attentions,
|
755 |
-
use_cache=use_cache,
|
756 |
-
slope_rate=slope_rate,
|
757 |
-
)
|
758 |
|
759 |
hidden_states = layer_outputs[0]
|
760 |
|
761 |
if use_cache:
|
762 |
-
next_decoder_cache += (
|
763 |
-
layer_outputs[2 if output_attentions else 1], )
|
764 |
|
765 |
if output_attentions:
|
766 |
-
all_self_attns += (layer_outputs[1],
|
767 |
|
768 |
hidden_states = self.final_norm(hidden_states)
|
769 |
|
770 |
# add hidden states from the last decoder layer
|
771 |
if output_hidden_states:
|
772 |
-
all_hidden_states += (hidden_states,
|
773 |
|
774 |
next_cache = next_decoder_cache if use_cache else None
|
775 |
if not return_dict:
|
776 |
return tuple(
|
777 |
-
v
|
778 |
-
[hidden_states, next_cache, all_hidden_states, all_self_attns]
|
779 |
-
if v is not None
|
|
|
780 |
return BaseModelOutputWithPast(
|
781 |
last_hidden_state=hidden_states,
|
782 |
past_key_values=next_cache,
|
@@ -786,7 +772,6 @@ class TransnormerModel(TransnormerPreTrainedModel):
|
|
786 |
|
787 |
|
788 |
class TransnormerForCausalLM(TransnormerPreTrainedModel):
|
789 |
-
|
790 |
def __init__(self, config):
|
791 |
super().__init__(config)
|
792 |
self.model = TransnormerModel(config)
|
@@ -794,9 +779,9 @@ class TransnormerForCausalLM(TransnormerPreTrainedModel):
|
|
794 |
logging_info(self.model)
|
795 |
|
796 |
# the lm_head weight is automatically tied to the embed tokens weight
|
797 |
-
self.lm_head = nn.Linear(
|
798 |
-
|
799 |
-
|
800 |
|
801 |
# Initialize weights and apply final processing
|
802 |
self.post_init()
|
@@ -820,8 +805,9 @@ class TransnormerForCausalLM(TransnormerPreTrainedModel):
|
|
820 |
return self.model
|
821 |
|
822 |
@add_start_docstrings_to_model_forward(TRANSNORMER_INPUTS_DOCSTRING)
|
823 |
-
@replace_return_docstrings(
|
824 |
-
|
|
|
825 |
def forward(
|
826 |
self,
|
827 |
input_ids: torch.LongTensor = None,
|
@@ -859,13 +845,19 @@ class TransnormerForCausalLM(TransnormerPreTrainedModel):
|
|
859 |
>>> tokenizer.batch_decode(generate_ids, skip_special_tokens=True, clean_up_tokenization_spaces=False)[0]
|
860 |
"Hey, are you consciours? Can you talk to me?\nI'm not consciours, but I can talk to you."
|
861 |
```"""
|
862 |
-
output_attentions = (
|
863 |
-
|
864 |
-
|
865 |
-
|
866 |
-
|
867 |
-
|
868 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
869 |
|
870 |
# decoder outputs consists of (dec_features, layer_state, dec_hidden, dec_attn)
|
871 |
outputs = self.model(
|
@@ -896,8 +888,8 @@ class TransnormerForCausalLM(TransnormerPreTrainedModel):
|
|
896 |
loss = loss_fct(shift_logits, shift_labels)
|
897 |
|
898 |
if not return_dict:
|
899 |
-
output = (logits,
|
900 |
-
return (loss,
|
901 |
|
902 |
return CausalLMOutputWithPast(
|
903 |
loss=loss,
|
@@ -924,149 +916,23 @@ class TransnormerForCausalLM(TransnormerPreTrainedModel):
|
|
924 |
else:
|
925 |
model_inputs = {"input_ids": input_ids}
|
926 |
|
927 |
-
model_inputs.update(
|
928 |
-
|
929 |
-
|
930 |
-
|
931 |
-
|
|
|
|
|
932 |
return model_inputs
|
933 |
|
934 |
@staticmethod
|
935 |
def _reorder_cache(past_key_values, beam_idx):
|
936 |
reordered_past = ()
|
937 |
for layer_past in past_key_values:
|
938 |
-
reordered_past += (
|
939 |
-
|
940 |
-
|
941 |
-
|
942 |
-
|
943 |
-
|
944 |
-
@add_start_docstrings(
|
945 |
-
"""
|
946 |
-
The LLaMa Model transformer with a sequence classification head on top (linear layer).
|
947 |
-
|
948 |
-
[`TransnormerForSequenceClassification`] uses the last token in order to do the classification, as other causal models
|
949 |
-
(e.g. GPT-2) do.
|
950 |
-
|
951 |
-
Since it does classification on the last token, it requires to know the position of the last token. If a
|
952 |
-
`pad_token_id` is defined in the configuration, it finds the last token that is not a padding token in each row. If
|
953 |
-
no `pad_token_id` is defined, it simply takes the last value in each row of the batch. Since it cannot guess the
|
954 |
-
padding tokens when `inputs_embeds` are passed instead of `input_ids`, it does the same (take the last value in
|
955 |
-
each row of the batch).
|
956 |
-
""",
|
957 |
-
TRANSNORMER_START_DOCSTRING,
|
958 |
-
)
|
959 |
-
class TransnormerForSequenceClassification(TransnormerPreTrainedModel):
|
960 |
-
_keys_to_ignore_on_load_missing = [r"lm_head.weight"]
|
961 |
-
|
962 |
-
def __init__(self, config):
|
963 |
-
super().__init__(config)
|
964 |
-
self.num_labels = config.num_labels
|
965 |
-
self.model = TransnormerModel(config)
|
966 |
-
self.score = nn.Linear(config.decoder_embed_dim,
|
967 |
-
self.num_labels,
|
968 |
-
bias=False)
|
969 |
-
|
970 |
-
# Initialize weights and apply final processing
|
971 |
-
self.post_init()
|
972 |
-
|
973 |
-
def get_input_embeddings(self):
|
974 |
-
return self.model.embed_tokens
|
975 |
-
|
976 |
-
def set_input_embeddings(self, value):
|
977 |
-
self.model.embed_tokens = value
|
978 |
-
|
979 |
-
@add_start_docstrings_to_model_forward(TRANSNORMER_INPUTS_DOCSTRING)
|
980 |
-
def forward(
|
981 |
-
self,
|
982 |
-
input_ids: torch.LongTensor = None,
|
983 |
-
attn_mask: Optional[torch.Tensor] = None,
|
984 |
-
past_key_values: Optional[List[torch.FloatTensor]] = None,
|
985 |
-
inputs_embeds: Optional[torch.FloatTensor] = None,
|
986 |
-
labels: Optional[torch.LongTensor] = None,
|
987 |
-
use_cache: Optional[bool] = None,
|
988 |
-
output_attentions: Optional[bool] = None,
|
989 |
-
output_hidden_states: Optional[bool] = None,
|
990 |
-
return_dict: Optional[bool] = None,
|
991 |
-
) -> Union[Tuple, SequenceClassifierOutputWithPast]:
|
992 |
-
r"""
|
993 |
-
labels (`torch.LongTensor` of shape `(batch_size,)`, *optional*):
|
994 |
-
Labels for computing the sequence classification/regression loss. Indices should be in `[0, ...,
|
995 |
-
config.num_labels - 1]`. If `config.num_labels == 1` a regression loss is computed (Mean-Square loss), If
|
996 |
-
`config.num_labels > 1` a classification loss is computed (Cross-Entropy).
|
997 |
-
"""
|
998 |
-
return_dict = (return_dict if return_dict is not None else
|
999 |
-
self.config.use_return_dict)
|
1000 |
-
|
1001 |
-
transformer_outputs = self.model(
|
1002 |
-
input_ids,
|
1003 |
-
attn_padding_mask=attn_mask,
|
1004 |
-
past_key_values=past_key_values,
|
1005 |
-
inputs_embeds=inputs_embeds,
|
1006 |
-
use_cache=use_cache,
|
1007 |
-
output_attentions=output_attentions,
|
1008 |
-
output_hidden_states=output_hidden_states,
|
1009 |
-
return_dict=return_dict,
|
1010 |
-
)
|
1011 |
-
hidden_states = transformer_outputs[0]
|
1012 |
-
|
1013 |
-
logits = self.score(hidden_states)
|
1014 |
-
|
1015 |
-
if input_ids is not None:
|
1016 |
-
batch_size = input_ids.shape[0]
|
1017 |
-
else:
|
1018 |
-
batch_size = inputs_embeds.shape[0]
|
1019 |
-
|
1020 |
-
if self.config.pad_token_id is None and batch_size != 1:
|
1021 |
-
raise ValueError(
|
1022 |
-
"Cannot handle batch sizes > 1 if no padding token is defined."
|
1023 |
)
|
1024 |
-
|
1025 |
-
sequence_lengths = -1
|
1026 |
-
else:
|
1027 |
-
if input_ids is not None:
|
1028 |
-
sequence_lengths = (
|
1029 |
-
torch.ne(input_ids, self.config.pad_token_id).sum(-1) -
|
1030 |
-
1).to(logits.device)
|
1031 |
-
else:
|
1032 |
-
sequence_lengths = -1
|
1033 |
-
|
1034 |
-
pooled_logits = logits[torch.arange(batch_size, device=logits.device),
|
1035 |
-
sequence_lengths]
|
1036 |
-
|
1037 |
-
loss = None
|
1038 |
-
if labels is not None:
|
1039 |
-
labels = labels.to(logits.device)
|
1040 |
-
if self.config.problem_type is None:
|
1041 |
-
if self.num_labels == 1:
|
1042 |
-
self.config.problem_type = "regression"
|
1043 |
-
elif self.num_labels > 1 and (labels.dtype == torch.long
|
1044 |
-
or labels.dtype == torch.int):
|
1045 |
-
self.config.problem_type = "single_label_classification"
|
1046 |
-
else:
|
1047 |
-
self.config.problem_type = "multi_label_classification"
|
1048 |
-
|
1049 |
-
if self.config.problem_type == "regression":
|
1050 |
-
loss_fct = MSELoss()
|
1051 |
-
if self.num_labels == 1:
|
1052 |
-
loss = loss_fct(pooled_logits.squeeze(), labels.squeeze())
|
1053 |
-
else:
|
1054 |
-
loss = loss_fct(pooled_logits, labels)
|
1055 |
-
elif self.config.problem_type == "single_label_classification":
|
1056 |
-
loss_fct = CrossEntropyLoss()
|
1057 |
-
loss = loss_fct(pooled_logits.view(-1, self.num_labels),
|
1058 |
-
labels.view(-1))
|
1059 |
-
elif self.config.problem_type == "multi_label_classification":
|
1060 |
-
loss_fct = BCEWithLogitsLoss()
|
1061 |
-
loss = loss_fct(pooled_logits, labels)
|
1062 |
-
if not return_dict:
|
1063 |
-
output = (pooled_logits, ) + transformer_outputs[1:]
|
1064 |
-
return ((loss, ) + output) if loss is not None else output
|
1065 |
|
1066 |
-
return SequenceClassifierOutputWithPast(
|
1067 |
-
loss=loss,
|
1068 |
-
logits=pooled_logits,
|
1069 |
-
past_key_values=transformer_outputs.past_key_values,
|
1070 |
-
hidden_states=transformer_outputs.hidden_states,
|
1071 |
-
attentions=transformer_outputs.attentions,
|
1072 |
-
)
|
|
|
11 |
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
12 |
# See the License for the specific language governing permissions and
|
13 |
# limitations under the License.
|
14 |
+
# coding=utf-8
|
|
|
15 |
""" PyTorch Transnormer model."""
|
16 |
import math
|
17 |
import os
|
|
|
28 |
from transformers.modeling_outputs import (
|
29 |
BaseModelOutputWithPast,
|
30 |
CausalLMOutputWithPast,
|
|
|
31 |
)
|
32 |
from transformers.modeling_utils import PreTrainedModel
|
33 |
from transformers.utils import (
|
|
|
83 |
########## start Transnormer
|
84 |
##### Linearized Relative Positional Encoding: https://openreview.net/forum?id=xoLyps2qWc&referrer=%5BAuthor%20Console%5D(%2Fgroup%3Fid%3DTMLR%2FAuthors%23your-submissions)
|
85 |
class Lrpe(nn.Module):
|
|
|
86 |
def __init__(
|
87 |
self,
|
88 |
num_heads=8,
|
|
|
92 |
d = num_heads * embed_dim
|
93 |
|
94 |
self.index = torch.empty(0)
|
95 |
+
self.theta = nn.Parameter(
|
96 |
+
10000 ** (-2 / d * torch.arange(d)).reshape(num_heads, 1, -1)
|
97 |
+
)
|
98 |
|
99 |
def extra_repr(self):
|
100 |
return print_module(self)
|
|
|
113 |
|
114 |
|
115 |
class GLU(nn.Module):
|
|
|
116 |
def __init__(self, d1, d2, bias=False):
|
117 |
super().__init__()
|
118 |
if debug:
|
|
|
135 |
|
136 |
|
137 |
class NormLinearAttention(nn.Module):
|
|
|
138 |
def __init__(
|
139 |
self,
|
140 |
embed_dim,
|
|
|
190 |
output_attentions,
|
191 |
past_key_value,
|
192 |
use_cache,
|
193 |
+
slope_rate,
|
194 |
)
|
195 |
# x: b n d
|
196 |
n = x.shape[-2]
|
|
|
198 |
q, k, v, u = self.qkvu_proj(x).chunk(4, dim=-1)
|
199 |
# reshape
|
200 |
q, k, v = map(
|
201 |
+
lambda x: rearrange(x, "b n (h d) -> b h n d", h=self.num_heads), [q, k, v]
|
202 |
+
)
|
203 |
# act
|
204 |
q = self.act(q)
|
205 |
k = self.act(k)
|
|
|
207 |
q_offset = 0
|
208 |
# lrpe relys on position, get cache first
|
209 |
if past_key_value is not None:
|
210 |
+
# reuse k, v, for evaluation only
|
211 |
k = torch.cat([past_key_value[0], k], dim=-2)
|
212 |
v = torch.cat([past_key_value[1], v], dim=-2)
|
213 |
q_offset = past_key_value[0].shape[-2]
|
|
|
224 |
|
225 |
if attn_padding_mask is not None:
|
226 |
v = v.masked_fill(
|
227 |
+
(1 - attn_padding_mask).unsqueeze(1).unsqueeze(-1).to(torch.bool), 0
|
228 |
+
)
|
229 |
|
230 |
if not has_lightning_attention:
|
231 |
if slope_rate != None:
|
232 |
attn_mask = torch.exp(slope_rate * attn_mask)
|
|
|
233 |
output = linear_attention(q, k, v, attn_mask)
|
234 |
else:
|
235 |
+
output = lightning_attention(
|
236 |
+
q, k, v, True, slope_rate.squeeze(-1).squeeze(-1)
|
237 |
+
)
|
238 |
|
239 |
# reshape
|
240 |
output = rearrange(output, "b h n d -> b n (h d)")
|
|
|
253 |
return output, attn_weights, past_key_value
|
254 |
|
255 |
def inference(
|
256 |
+
self,
|
257 |
+
x,
|
258 |
+
attn_mask: Optional[torch.Tensor] = None, # (b, h, n, m)
|
259 |
+
attn_padding_mask: Optional[torch.Tensor] = None, # (b, m)
|
260 |
+
output_attentions: bool = False,
|
261 |
+
past_key_value: Optional[Tuple[torch.Tensor]] = None,
|
262 |
+
use_cache: bool = False,
|
263 |
+
slope_rate: Optional[torch.Tensor] = None, # (h, 1, 1)
|
264 |
):
|
265 |
# x: b n d
|
266 |
n = x.shape[-2]
|
|
|
268 |
q, k, v, u = self.qkvu_proj(x).chunk(4, dim=-1)
|
269 |
# reshape
|
270 |
q, k, v = map(
|
271 |
+
lambda x: rearrange(x, "b n (h d) -> b h n d", h=self.num_heads), [q, k, v]
|
272 |
+
)
|
273 |
# act
|
274 |
q = self.act(q)
|
275 |
k = self.act(k)
|
|
|
277 |
# rpe
|
278 |
if self.linear_use_lrpe:
|
279 |
q = self.lrpe(q, offset=self.offset)
|
280 |
+
k = self.lrpe(k)
|
281 |
|
282 |
if past_key_value == None:
|
283 |
self.offset = q.shape[-2]
|
|
|
295 |
|
296 |
if attn_padding_mask is not None:
|
297 |
attn_mask = attn_mask.masked_fill(
|
298 |
+
(1 - attn_padding_mask).unsqueeze(1).unsqueeze(2).to(torch.bool),
|
|
|
299 |
0,
|
300 |
)
|
301 |
energy = torch.einsum("... n d, ... m d -> ... n m", q, k)
|
|
|
306 |
output = torch.einsum("... n m, ... m d -> ... n d", energy, v)
|
307 |
|
308 |
eval_and_not_generate = eval(
|
309 |
+
os.environ.get("eval_and_not_generate", default="False")
|
310 |
+
)
|
311 |
if eval_and_not_generate:
|
312 |
kv = None
|
313 |
else:
|
314 |
# b, h, n, e, d
|
315 |
+
kv_outproduct = torch.einsum("... n e, ... n d -> ... n e d", k, v)
|
|
|
316 |
# 1, 1, n, 1, 1
|
317 |
+
index = torch.arange(n - 1, -1, -1).reshape(1, 1, -1, 1, 1).to(x)
|
|
|
318 |
# (h, 1, 1) -> (1, h, 1, 1, 1); (1, h, 1, 1, 1), (1, 1, n, 1, 1) -> (1, h, n, 1, 1)
|
319 |
+
decay = ratio.unsqueeze(0).unsqueeze(-1) ** index
|
320 |
|
321 |
kv_outproduct_with_decay = kv_outproduct * decay
|
322 |
kv = torch.sum(kv_outproduct_with_decay, dim=-3)
|
|
|
327 |
for i in range(n):
|
328 |
kv = ratio * kv + torch.einsum(
|
329 |
"... n d, ... n e -> ... d e",
|
330 |
+
k[:, :, i : i + 1],
|
331 |
+
v[:, :, i : i + 1],
|
332 |
+
)
|
333 |
+
qkv = torch.einsum(
|
334 |
+
"... n e, ... e d -> ... n d", q[:, :, i : i + 1], kv
|
335 |
)
|
|
|
|
|
336 |
output.append(qkv)
|
337 |
output = torch.concat(output, dim=-2)
|
338 |
|
|
|
351 |
|
352 |
|
353 |
class TransnormerDecoderLayer(nn.Module):
|
|
|
354 |
def __init__(self, config: TransnormerConfig):
|
355 |
super().__init__()
|
356 |
self.embed_dim = config.decoder_embed_dim
|
|
|
389 |
return residual + x
|
390 |
|
391 |
def forward(
|
392 |
+
self,
|
393 |
+
x,
|
394 |
+
attn_mask: Optional[torch.Tensor] = None,
|
395 |
+
attn_padding_mask: Optional[torch.Tensor] = None,
|
396 |
+
past_key_value: Optional[Tuple[torch.Tensor]] = None,
|
397 |
+
output_attentions: Optional[bool] = False,
|
398 |
+
use_cache: Optional[bool] = False,
|
399 |
+
slope_rate: Optional[torch.Tensor] = None, # (h, 1, 1)
|
400 |
):
|
401 |
residual = x
|
402 |
x = self.token_norm(x)
|
|
|
416 |
x = self.channel_mixer(x)
|
417 |
x = self.residual_connection(x, residual)
|
418 |
|
419 |
+
outputs = (x,)
|
420 |
|
421 |
if output_attentions:
|
422 |
+
outputs += (self_attn_weights,)
|
423 |
|
424 |
if use_cache:
|
425 |
+
outputs += (present_key_value,)
|
426 |
|
427 |
return outputs
|
428 |
|
|
|
444 |
"""
|
445 |
|
446 |
|
447 |
+
@add_start_docstrings(
|
448 |
+
TRANSNORMER_START_DOCSTRING,
|
449 |
+
)
|
450 |
class TransnormerPreTrainedModel(PreTrainedModel):
|
451 |
config_class = TransnormerConfig
|
452 |
base_model_prefix = "model"
|
|
|
531 |
"""
|
532 |
|
533 |
|
534 |
+
@add_start_docstrings(
|
535 |
+
TRANSNORMER_START_DOCSTRING,
|
536 |
+
)
|
537 |
class TransnormerModel(TransnormerPreTrainedModel):
|
538 |
"""
|
539 |
Transformer decoder consisting of *config.num_hidden_layers* layers. Each layer is a [`TransnormerDecoderLayer`]
|
|
|
557 |
self.slopes = self._build_slope_tensor(config.decoder_attention_heads)
|
558 |
|
559 |
# params
|
560 |
+
self.embed_tokens = nn.Embedding(
|
561 |
+
config.vocab_size, config.decoder_embed_dim, self.padding_idx
|
562 |
+
)
|
563 |
self.layers = nn.ModuleList([])
|
564 |
for i in range(config.decoder_layers):
|
565 |
if len(self.linear_use_lrpe_list) > 0:
|
566 |
config.linear_use_lrpe = self.linear_use_lrpe_list[i]
|
567 |
self.layers.append(TransnormerDecoderLayer(config))
|
568 |
|
569 |
+
self.final_norm = get_norm_fn(config.norm_type)(config.decoder_embed_dim)
|
|
|
570 |
self.embed_dim = config.decoder_embed_dim
|
571 |
+
self.embed_scale = (
|
572 |
+
1.0 if config.no_scale_embedding else math.sqrt(self.embed_dim)
|
573 |
+
)
|
574 |
|
575 |
# Initialize weights and apply final processing
|
576 |
self.post_init()
|
577 |
|
578 |
@staticmethod
|
579 |
def _build_slope_tensor(n_attention_heads: int):
|
|
|
580 |
def get_slopes(n):
|
|
|
581 |
def get_slopes_power_of_2(n):
|
582 |
+
start = 2 ** (-(2 ** -(math.log2(n) - 3)))
|
583 |
ratio = start
|
584 |
return [start * ratio**i for i in range(n)]
|
585 |
|
|
|
588 |
n
|
589 |
) # In the paper, we only train models that have 2^a heads for some a. This function has
|
590 |
else: # some good properties that only occur when the input is a power of 2. To maintain that even
|
591 |
+
closest_power_of_2 = 2 ** math.floor(
|
592 |
math.log2(n)
|
593 |
) # when the number of heads is not a power of 2, we use this workaround.
|
594 |
+
return (
|
595 |
+
get_slopes_power_of_2(closest_power_of_2)
|
596 |
+
+ get_slopes(2 * closest_power_of_2)[0::2][: n - closest_power_of_2]
|
597 |
+
)
|
598 |
|
599 |
# h, 1, 1
|
600 |
slopes = torch.tensor(get_slopes(n_attention_heads)).reshape(
|
601 |
+
n_attention_heads, 1, 1
|
602 |
+
)
|
603 |
|
604 |
return slopes
|
605 |
|
|
|
612 |
def set_input_embeddings(self, value):
|
613 |
self.embed_tokens = value
|
614 |
|
615 |
+
def _prepare_decoder_linear_attn_mask(
|
616 |
+
self, input_shape, inputs_embeds, past_key_values_length
|
617 |
+
):
|
618 |
bsz, tgt_len = input_shape
|
619 |
src_len = tgt_len + past_key_values_length
|
620 |
|
621 |
def power_log(x):
|
622 |
+
return 2 ** (math.ceil(math.log(x, 2)))
|
623 |
|
624 |
n = power_log(max(tgt_len, src_len))
|
625 |
if self._linear_attn_mask.shape[-1] < n:
|
626 |
|
627 |
def get_mask(n):
|
628 |
+
mask = torch.triu(torch.zeros(n, n).float().fill_(float("-inf")), 1)
|
|
|
629 |
# no slope version
|
630 |
# -n, ..., -2, -1, 0
|
631 |
for i in range(n):
|
632 |
x = torch.arange(i + 1)
|
633 |
y = x
|
634 |
+
mask[i, : i + 1] = -torch.flip(y, [0])
|
635 |
|
636 |
return mask
|
637 |
|
|
|
643 |
linear_attn_mask = self._linear_attn_mask[:, -tgt_len:, -src_len:]
|
644 |
num_heads = linear_attn_mask.shape[0]
|
645 |
|
646 |
+
return linear_attn_mask[None, :, :, :].expand(bsz, num_heads, tgt_len, src_len)
|
|
|
647 |
|
648 |
@add_start_docstrings_to_model_forward(TRANSNORMER_INPUTS_DOCSTRING)
|
649 |
def forward(
|
|
|
657 |
output_hidden_states: Optional[bool] = None,
|
658 |
return_dict: Optional[bool] = None,
|
659 |
) -> Union[Tuple, BaseModelOutputWithPast]:
|
660 |
+
output_attentions = (
|
661 |
+
output_attentions
|
662 |
+
if output_attentions is not None
|
663 |
+
else self.config.output_attentions
|
664 |
+
)
|
665 |
+
output_hidden_states = (
|
666 |
+
output_hidden_states
|
667 |
+
if output_hidden_states is not None
|
668 |
+
else self.config.output_hidden_states
|
669 |
+
)
|
670 |
use_cache = use_cache if use_cache is not None else self.config.use_cache
|
671 |
|
672 |
+
return_dict = (
|
673 |
+
return_dict if return_dict is not None else self.config.use_return_dict
|
674 |
+
)
|
675 |
|
676 |
# retrieve input_ids and inputs_embeds
|
677 |
if input_ids is not None and inputs_embeds is not None:
|
|
|
693 |
if past_key_values is not None:
|
694 |
past_key_values_length = past_key_values[0][0].shape[-2]
|
695 |
seq_length_with_past = seq_length_with_past + past_key_values_length
|
696 |
+
|
697 |
if inputs_embeds is None:
|
698 |
# !!! use embed_scale
|
699 |
inputs_embeds = self.embed_scale * self.embed_tokens(input_ids)
|
|
|
715 |
##### norm linear layers
|
716 |
linear_attn_padding_mask = attn_padding_mask
|
717 |
linear_attn_mask = self._prepare_decoder_linear_attn_mask(
|
718 |
+
(batch_size, seq_length), inputs_embeds, past_key_values_length
|
719 |
+
)
|
720 |
|
721 |
+
slope_rates = [self.slopes.to(input_ids.device) for _ in range(self.num_layers)]
|
|
|
|
|
722 |
|
723 |
for idx, layer in enumerate(self.layers):
|
724 |
if output_hidden_states:
|
725 |
+
all_hidden_states += (hidden_states,)
|
726 |
|
727 |
+
past_key_value = (
|
728 |
+
past_key_values[idx] if past_key_values is not None else None
|
729 |
+
)
|
730 |
|
731 |
slope_rate = slope_rates[idx]
|
732 |
slope_rate = slope_rate * (1 - idx / (self.num_layers - 1) + 1e-5)
|
733 |
mask = linear_attn_mask
|
734 |
+
|
735 |
+
layer_outputs = layer(
|
736 |
+
hidden_states,
|
737 |
+
attn_mask=mask,
|
738 |
+
attn_padding_mask=linear_attn_padding_mask,
|
739 |
+
past_key_value=past_key_value,
|
740 |
+
output_attentions=output_attentions,
|
741 |
+
use_cache=use_cache,
|
742 |
+
slope_rate=slope_rate,
|
743 |
+
)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
744 |
|
745 |
hidden_states = layer_outputs[0]
|
746 |
|
747 |
if use_cache:
|
748 |
+
next_decoder_cache += (layer_outputs[2 if output_attentions else 1],)
|
|
|
749 |
|
750 |
if output_attentions:
|
751 |
+
all_self_attns += (layer_outputs[1],)
|
752 |
|
753 |
hidden_states = self.final_norm(hidden_states)
|
754 |
|
755 |
# add hidden states from the last decoder layer
|
756 |
if output_hidden_states:
|
757 |
+
all_hidden_states += (hidden_states,)
|
758 |
|
759 |
next_cache = next_decoder_cache if use_cache else None
|
760 |
if not return_dict:
|
761 |
return tuple(
|
762 |
+
v
|
763 |
+
for v in [hidden_states, next_cache, all_hidden_states, all_self_attns]
|
764 |
+
if v is not None
|
765 |
+
)
|
766 |
return BaseModelOutputWithPast(
|
767 |
last_hidden_state=hidden_states,
|
768 |
past_key_values=next_cache,
|
|
|
772 |
|
773 |
|
774 |
class TransnormerForCausalLM(TransnormerPreTrainedModel):
|
|
|
775 |
def __init__(self, config):
|
776 |
super().__init__(config)
|
777 |
self.model = TransnormerModel(config)
|
|
|
779 |
logging_info(self.model)
|
780 |
|
781 |
# the lm_head weight is automatically tied to the embed tokens weight
|
782 |
+
self.lm_head = nn.Linear(
|
783 |
+
config.decoder_embed_dim, config.vocab_size, bias=False
|
784 |
+
)
|
785 |
|
786 |
# Initialize weights and apply final processing
|
787 |
self.post_init()
|
|
|
805 |
return self.model
|
806 |
|
807 |
@add_start_docstrings_to_model_forward(TRANSNORMER_INPUTS_DOCSTRING)
|
808 |
+
@replace_return_docstrings(
|
809 |
+
output_type=CausalLMOutputWithPast, config_class=_CONFIG_FOR_DOC
|
810 |
+
)
|
811 |
def forward(
|
812 |
self,
|
813 |
input_ids: torch.LongTensor = None,
|
|
|
845 |
>>> tokenizer.batch_decode(generate_ids, skip_special_tokens=True, clean_up_tokenization_spaces=False)[0]
|
846 |
"Hey, are you consciours? Can you talk to me?\nI'm not consciours, but I can talk to you."
|
847 |
```"""
|
848 |
+
output_attentions = (
|
849 |
+
output_attentions
|
850 |
+
if output_attentions is not None
|
851 |
+
else self.config.output_attentions
|
852 |
+
)
|
853 |
+
output_hidden_states = (
|
854 |
+
output_hidden_states
|
855 |
+
if output_hidden_states is not None
|
856 |
+
else self.config.output_hidden_states
|
857 |
+
)
|
858 |
+
return_dict = (
|
859 |
+
return_dict if return_dict is not None else self.config.use_return_dict
|
860 |
+
)
|
861 |
|
862 |
# decoder outputs consists of (dec_features, layer_state, dec_hidden, dec_attn)
|
863 |
outputs = self.model(
|
|
|
888 |
loss = loss_fct(shift_logits, shift_labels)
|
889 |
|
890 |
if not return_dict:
|
891 |
+
output = (logits,) + outputs[1:]
|
892 |
+
return (loss,) + output if loss is not None else output
|
893 |
|
894 |
return CausalLMOutputWithPast(
|
895 |
loss=loss,
|
|
|
916 |
else:
|
917 |
model_inputs = {"input_ids": input_ids}
|
918 |
|
919 |
+
model_inputs.update(
|
920 |
+
{
|
921 |
+
"past_key_values": past_key_values,
|
922 |
+
"use_cache": kwargs.get("use_cache"),
|
923 |
+
"attention_mask": attention_mask,
|
924 |
+
}
|
925 |
+
)
|
926 |
return model_inputs
|
927 |
|
928 |
@staticmethod
|
929 |
def _reorder_cache(past_key_values, beam_idx):
|
930 |
reordered_past = ()
|
931 |
for layer_past in past_key_values:
|
932 |
+
reordered_past += (
|
933 |
+
tuple(
|
934 |
+
past_state.index_select(0, beam_idx) for past_state in layer_past
|
935 |
+
),
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
936 |
)
|
937 |
+
return reordered_past
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
938 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
tokenization_baichuan.py
CHANGED
@@ -73,6 +73,11 @@ class BaiChuanTokenizer(PreTrainedTokenizer):
|
|
73 |
if isinstance(unk_token, str) else unk_token)
|
74 |
pad_token = (AddedToken(pad_token, lstrip=False, rstrip=False)
|
75 |
if isinstance(pad_token, str) else pad_token)
|
|
|
|
|
|
|
|
|
|
|
76 |
super().__init__(
|
77 |
bos_token=bos_token,
|
78 |
eos_token=eos_token,
|
@@ -84,11 +89,6 @@ class BaiChuanTokenizer(PreTrainedTokenizer):
|
|
84 |
clean_up_tokenization_spaces=clean_up_tokenization_spaces,
|
85 |
**kwargs,
|
86 |
)
|
87 |
-
self.vocab_file = vocab_file
|
88 |
-
self.add_bos_token = add_bos_token
|
89 |
-
self.add_eos_token = add_eos_token
|
90 |
-
self.sp_model = spm.SentencePieceProcessor(**self.sp_model_kwargs)
|
91 |
-
self.sp_model.Load(vocab_file)
|
92 |
|
93 |
def __getstate__(self):
|
94 |
state = self.__dict__.copy()
|
|
|
73 |
if isinstance(unk_token, str) else unk_token)
|
74 |
pad_token = (AddedToken(pad_token, lstrip=False, rstrip=False)
|
75 |
if isinstance(pad_token, str) else pad_token)
|
76 |
+
self.vocab_file = vocab_file
|
77 |
+
self.add_bos_token = add_bos_token
|
78 |
+
self.add_eos_token = add_eos_token
|
79 |
+
self.sp_model = spm.SentencePieceProcessor(**self.sp_model_kwargs)
|
80 |
+
self.sp_model.Load(vocab_file)
|
81 |
super().__init__(
|
82 |
bos_token=bos_token,
|
83 |
eos_token=eos_token,
|
|
|
89 |
clean_up_tokenization_spaces=clean_up_tokenization_spaces,
|
90 |
**kwargs,
|
91 |
)
|
|
|
|
|
|
|
|
|
|
|
92 |
|
93 |
def __getstate__(self):
|
94 |
state = self.__dict__.copy()
|