Spaces:
Sleeping
Sleeping
Balaji S
commited on
Removed unnecessary import
Browse files
model.py
CHANGED
@@ -1,325 +1,324 @@
|
|
1 |
-
import math
|
2 |
-
import torch
|
3 |
-
import numpy as np
|
4 |
-
import torch.nn as nn
|
5 |
-
from tqdm import tqdm
|
6 |
-
import scipy.sparse as sp
|
7 |
-
import torch.nn.functional as F
|
8 |
-
import torch.distributed as dist
|
9 |
-
|
10 |
-
import transformers
|
11 |
-
from transformers import RobertaTokenizer
|
12 |
-
from transformers.models.roberta.modeling_roberta import RobertaPreTrainedModel, RobertaModel, RobertaLMHead
|
13 |
-
from transformers.models.bert.modeling_bert import BertPreTrainedModel, BertModel, BertLMPredictionHead
|
14 |
-
from transformers.activations import gelu
|
15 |
-
from transformers.file_utils import (
|
16 |
-
add_code_sample_docstrings,
|
17 |
-
add_start_docstrings,
|
18 |
-
add_start_docstrings_to_model_forward,
|
19 |
-
replace_return_docstrings,
|
20 |
-
)
|
21 |
-
from transformers.modeling_outputs import SequenceClassifierOutput, BaseModelOutputWithPoolingAndCrossAttentions
|
22 |
-
|
23 |
-
|
24 |
-
|
25 |
-
|
26 |
-
|
27 |
-
|
28 |
-
|
29 |
-
|
30 |
-
|
31 |
-
|
32 |
-
|
33 |
-
r
|
34 |
-
|
35 |
-
|
36 |
-
|
37 |
-
|
38 |
-
|
39 |
-
|
40 |
-
|
41 |
-
|
42 |
-
self.
|
43 |
-
|
44 |
-
|
45 |
-
|
46 |
-
x = self.
|
47 |
-
|
48 |
-
|
49 |
-
|
50 |
-
|
51 |
-
|
52 |
-
|
53 |
-
|
54 |
-
'
|
55 |
-
'
|
56 |
-
'
|
57 |
-
'
|
58 |
-
|
59 |
-
|
60 |
-
|
61 |
-
|
62 |
-
self.pooler_type
|
63 |
-
|
64 |
-
|
65 |
-
|
66 |
-
|
67 |
-
|
68 |
-
|
69 |
-
|
70 |
-
|
71 |
-
|
72 |
-
|
73 |
-
|
74 |
-
|
75 |
-
|
76 |
-
|
77 |
-
pooled_result
|
78 |
-
|
79 |
-
|
80 |
-
|
81 |
-
|
82 |
-
pooled_result
|
83 |
-
|
84 |
-
|
85 |
-
|
86 |
-
|
87 |
-
|
88 |
-
|
89 |
-
|
90 |
-
|
91 |
-
|
92 |
-
|
93 |
-
|
94 |
-
self.
|
95 |
-
|
96 |
-
|
97 |
-
|
98 |
-
|
99 |
-
|
100 |
-
|
101 |
-
|
102 |
-
|
103 |
-
|
104 |
-
|
105 |
-
|
106 |
-
|
107 |
-
self.
|
108 |
-
self.
|
109 |
-
|
110 |
-
|
111 |
-
|
112 |
-
|
113 |
-
|
114 |
-
|
115 |
-
|
116 |
-
self.
|
117 |
-
self.
|
118 |
-
self.
|
119 |
-
|
120 |
-
|
121 |
-
self.
|
122 |
-
self.
|
123 |
-
self.
|
124 |
-
self.
|
125 |
-
self.
|
126 |
-
|
127 |
-
|
128 |
-
|
129 |
-
|
130 |
-
|
131 |
-
|
132 |
-
|
133 |
-
|
134 |
-
|
135 |
-
|
136 |
-
|
137 |
-
|
138 |
-
|
139 |
-
|
140 |
-
|
141 |
-
|
142 |
-
|
143 |
-
|
144 |
-
|
145 |
-
|
146 |
-
|
147 |
-
|
148 |
-
|
149 |
-
|
150 |
-
|
151 |
-
|
152 |
-
|
153 |
-
|
154 |
-
|
155 |
-
|
156 |
-
|
157 |
-
|
158 |
-
|
159 |
-
|
160 |
-
|
161 |
-
|
162 |
-
|
163 |
-
|
164 |
-
|
165 |
-
|
166 |
-
|
167 |
-
|
168 |
-
|
169 |
-
|
170 |
-
|
171 |
-
|
172 |
-
|
173 |
-
|
174 |
-
|
175 |
-
|
176 |
-
|
177 |
-
|
178 |
-
|
179 |
-
|
180 |
-
|
181 |
-
|
182 |
-
|
183 |
-
|
184 |
-
|
185 |
-
|
186 |
-
|
187 |
-
|
188 |
-
|
189 |
-
|
190 |
-
|
191 |
-
|
192 |
-
|
193 |
-
|
194 |
-
|
195 |
-
|
196 |
-
|
197 |
-
|
198 |
-
|
199 |
-
|
200 |
-
|
201 |
-
|
202 |
-
|
203 |
-
|
204 |
-
|
205 |
-
|
206 |
-
|
207 |
-
|
208 |
-
|
209 |
-
|
210 |
-
|
211 |
-
#
|
212 |
-
|
213 |
-
|
214 |
-
|
215 |
-
|
216 |
-
|
217 |
-
|
218 |
-
|
219 |
-
|
220 |
-
|
221 |
-
|
222 |
-
|
223 |
-
|
224 |
-
|
225 |
-
dist.all_gather(tensor_list=
|
226 |
-
dist.all_gather(tensor_list=
|
227 |
-
|
228 |
-
|
229 |
-
#
|
230 |
-
|
231 |
-
|
232 |
-
|
233 |
-
|
234 |
-
|
235 |
-
|
236 |
-
|
237 |
-
|
238 |
-
|
239 |
-
|
240 |
-
|
241 |
-
|
242 |
-
|
243 |
-
|
244 |
-
|
245 |
-
|
246 |
-
|
247 |
-
|
248 |
-
|
249 |
-
|
250 |
-
|
251 |
-
|
252 |
-
|
253 |
-
|
254 |
-
|
255 |
-
|
256 |
-
|
257 |
-
|
258 |
-
|
259 |
-
|
260 |
-
|
261 |
-
|
262 |
-
|
263 |
-
|
264 |
-
|
265 |
-
|
266 |
-
|
267 |
-
|
268 |
-
|
269 |
-
|
270 |
-
|
271 |
-
|
272 |
-
|
273 |
-
|
274 |
-
|
275 |
-
|
276 |
-
|
277 |
-
|
278 |
-
|
279 |
-
|
280 |
-
|
281 |
-
|
282 |
-
|
283 |
-
|
284 |
-
|
285 |
-
|
286 |
-
|
287 |
-
)
|
288 |
-
|
289 |
-
|
290 |
-
|
291 |
-
|
292 |
-
|
293 |
-
|
294 |
-
|
295 |
-
|
296 |
-
|
297 |
-
|
298 |
-
|
299 |
-
|
300 |
-
|
301 |
-
|
302 |
-
|
303 |
-
|
304 |
-
|
305 |
-
|
306 |
-
|
307 |
-
|
308 |
-
|
309 |
-
|
310 |
-
|
311 |
-
|
312 |
-
|
313 |
-
|
314 |
-
|
315 |
-
|
316 |
-
|
317 |
-
|
318 |
-
|
319 |
-
|
320 |
-
|
321 |
-
|
322 |
-
|
323 |
-
|
324 |
-
|
325 |
-
return user_embeds, item_embeds
|
|
|
1 |
+
import math
|
2 |
+
import torch
|
3 |
+
import numpy as np
|
4 |
+
import torch.nn as nn
|
5 |
+
from tqdm import tqdm
|
6 |
+
import scipy.sparse as sp
|
7 |
+
import torch.nn.functional as F
|
8 |
+
import torch.distributed as dist
|
9 |
+
|
10 |
+
import transformers
|
11 |
+
from transformers import RobertaTokenizer
|
12 |
+
from transformers.models.roberta.modeling_roberta import RobertaPreTrainedModel, RobertaModel, RobertaLMHead
|
13 |
+
from transformers.models.bert.modeling_bert import BertPreTrainedModel, BertModel, BertLMPredictionHead
|
14 |
+
from transformers.activations import gelu
|
15 |
+
from transformers.file_utils import (
|
16 |
+
add_code_sample_docstrings,
|
17 |
+
add_start_docstrings,
|
18 |
+
add_start_docstrings_to_model_forward,
|
19 |
+
replace_return_docstrings,
|
20 |
+
)
|
21 |
+
from transformers.modeling_outputs import SequenceClassifierOutput, BaseModelOutputWithPoolingAndCrossAttentions
|
22 |
+
|
23 |
+
|
24 |
+
init = nn.init.xavier_uniform_
|
25 |
+
uniformInit = nn.init.uniform
|
26 |
+
|
27 |
+
|
28 |
+
"""
|
29 |
+
EasyRec
|
30 |
+
"""
|
31 |
+
def dot_product_scores(q_vectors, ctx_vectors):
|
32 |
+
r = torch.matmul(q_vectors, torch.transpose(ctx_vectors, 0, 1))
|
33 |
+
return r
|
34 |
+
|
35 |
+
class MLPLayer(nn.Module):
|
36 |
+
"""
|
37 |
+
Head for getting sentence representations over RoBERTa/BERT's CLS representation.
|
38 |
+
"""
|
39 |
+
def __init__(self, config):
|
40 |
+
super().__init__()
|
41 |
+
self.dense = nn.Linear(config.hidden_size, config.hidden_size)
|
42 |
+
self.activation = nn.Tanh()
|
43 |
+
|
44 |
+
def forward(self, features, **kwargs):
|
45 |
+
x = self.dense(features)
|
46 |
+
x = self.activation(x)
|
47 |
+
return x
|
48 |
+
|
49 |
+
|
50 |
+
class Pooler(nn.Module):
|
51 |
+
"""
|
52 |
+
Parameter-free poolers to get the sentence embedding
|
53 |
+
'cls': [CLS] representation with BERT/RoBERTa's MLP pooler.
|
54 |
+
'cls_before_pooler': [CLS] representation without the original MLP pooler.
|
55 |
+
'avg': average of the last layers' hidden states at each token.
|
56 |
+
'avg_top2': average of the last two layers.
|
57 |
+
'avg_first_last': average of the first and the last layers.
|
58 |
+
"""
|
59 |
+
def __init__(self, pooler_type):
|
60 |
+
super().__init__()
|
61 |
+
self.pooler_type = pooler_type
|
62 |
+
assert self.pooler_type in ["cls", "cls_before_pooler", "avg", "avg_top2", "avg_first_last"], "unrecognized pooling type %s" % self.pooler_type
|
63 |
+
|
64 |
+
def forward(self, attention_mask, outputs):
|
65 |
+
last_hidden = outputs.last_hidden_state
|
66 |
+
pooler_output = outputs.pooler_output
|
67 |
+
hidden_states = outputs.hidden_states
|
68 |
+
|
69 |
+
if self.pooler_type in ['cls_before_pooler', 'cls']:
|
70 |
+
return last_hidden[:, 0]
|
71 |
+
elif self.pooler_type == "avg":
|
72 |
+
return ((last_hidden * attention_mask.unsqueeze(-1)).sum(1) / attention_mask.sum(-1).unsqueeze(-1))
|
73 |
+
elif self.pooler_type == "avg_first_last":
|
74 |
+
first_hidden = hidden_states[1]
|
75 |
+
last_hidden = hidden_states[-1]
|
76 |
+
pooled_result = ((first_hidden + last_hidden) / 2.0 * attention_mask.unsqueeze(-1)).sum(1) / attention_mask.sum(-1).unsqueeze(-1)
|
77 |
+
return pooled_result
|
78 |
+
elif self.pooler_type == "avg_top2":
|
79 |
+
second_last_hidden = hidden_states[-2]
|
80 |
+
last_hidden = hidden_states[-1]
|
81 |
+
pooled_result = ((last_hidden + second_last_hidden) / 2.0 * attention_mask.unsqueeze(-1)).sum(1) / attention_mask.sum(-1).unsqueeze(-1)
|
82 |
+
return pooled_result
|
83 |
+
else:
|
84 |
+
raise NotImplementedError
|
85 |
+
|
86 |
+
|
87 |
+
class Similarity(nn.Module):
|
88 |
+
"""
|
89 |
+
Dot product or cosine similarity
|
90 |
+
"""
|
91 |
+
def __init__(self, temp):
|
92 |
+
super().__init__()
|
93 |
+
self.temp = temp
|
94 |
+
self.cos = nn.CosineSimilarity(dim=-1)
|
95 |
+
|
96 |
+
def forward(self, x, y):
|
97 |
+
return self.cos(x, y) / self.temp
|
98 |
+
|
99 |
+
|
100 |
+
class Easyrec(RobertaPreTrainedModel):
|
101 |
+
_keys_to_ignore_on_load_missing = [r"position_ids"]
|
102 |
+
|
103 |
+
def __init__(self, config, *model_args, **model_kargs):
|
104 |
+
super().__init__(config)
|
105 |
+
try:
|
106 |
+
self.model_args = model_kargs["model_args"]
|
107 |
+
self.roberta = RobertaModel(config, add_pooling_layer=False)
|
108 |
+
if self.model_args.pooler_type == "cls":
|
109 |
+
self.mlp = MLPLayer(config)
|
110 |
+
if self.model_args.do_mlm:
|
111 |
+
self.lm_head = RobertaLMHead(config)
|
112 |
+
"""
|
113 |
+
Contrastive learning class init function.
|
114 |
+
"""
|
115 |
+
self.pooler_type = self.model_args.pooler_type
|
116 |
+
self.pooler = Pooler(self.pooler_type)
|
117 |
+
self.sim = Similarity(temp=self.model_args.temp)
|
118 |
+
self.init_weights()
|
119 |
+
except:
|
120 |
+
self.roberta = RobertaModel(config, add_pooling_layer=False)
|
121 |
+
self.mlp = MLPLayer(config)
|
122 |
+
self.lm_head = RobertaLMHead(config)
|
123 |
+
self.pooler_type = 'cls'
|
124 |
+
self.pooler = Pooler(self.pooler_type)
|
125 |
+
self.init_weights()
|
126 |
+
|
127 |
+
def forward(self,
|
128 |
+
user_input_ids=None,
|
129 |
+
user_attention_mask=None,
|
130 |
+
pos_item_input_ids=None,
|
131 |
+
pos_item_attention_mask=None,
|
132 |
+
neg_item_input_ids=None,
|
133 |
+
neg_item_attention_mask=None,
|
134 |
+
token_type_ids=None,
|
135 |
+
position_ids=None,
|
136 |
+
head_mask=None,
|
137 |
+
inputs_embeds=None,
|
138 |
+
labels=None,
|
139 |
+
output_attentions=None,
|
140 |
+
output_hidden_states=None,
|
141 |
+
return_dict=None,
|
142 |
+
mlm_input_ids=None,
|
143 |
+
mlm_attention_mask=None,
|
144 |
+
mlm_labels=None,
|
145 |
+
):
|
146 |
+
"""
|
147 |
+
Contrastive learning forward function.
|
148 |
+
"""
|
149 |
+
return_dict = return_dict if return_dict is not None else self.config.use_return_dict
|
150 |
+
batch_size = user_input_ids.size(0)
|
151 |
+
|
152 |
+
# Get user embeddings
|
153 |
+
user_outputs = self.roberta(
|
154 |
+
input_ids=user_input_ids,
|
155 |
+
attention_mask=user_attention_mask,
|
156 |
+
token_type_ids=None,
|
157 |
+
position_ids=None,
|
158 |
+
head_mask=None,
|
159 |
+
inputs_embeds=None,
|
160 |
+
output_attentions=output_attentions,
|
161 |
+
output_hidden_states=output_hidden_states,
|
162 |
+
return_dict=return_dict,
|
163 |
+
)
|
164 |
+
|
165 |
+
# Get positive item embeddings
|
166 |
+
pos_item_outputs = self.roberta(
|
167 |
+
input_ids=pos_item_input_ids,
|
168 |
+
attention_mask=pos_item_attention_mask,
|
169 |
+
token_type_ids=None,
|
170 |
+
position_ids=None,
|
171 |
+
head_mask=None,
|
172 |
+
inputs_embeds=None,
|
173 |
+
output_attentions=output_attentions,
|
174 |
+
output_hidden_states=output_hidden_states,
|
175 |
+
return_dict=return_dict,
|
176 |
+
)
|
177 |
+
|
178 |
+
# Get negative item embeddings
|
179 |
+
neg_item_outputs = self.roberta(
|
180 |
+
input_ids=neg_item_input_ids,
|
181 |
+
attention_mask=neg_item_attention_mask,
|
182 |
+
token_type_ids=None,
|
183 |
+
position_ids=None,
|
184 |
+
head_mask=None,
|
185 |
+
inputs_embeds=None,
|
186 |
+
output_attentions=output_attentions,
|
187 |
+
output_hidden_states=output_hidden_states,
|
188 |
+
return_dict=return_dict,
|
189 |
+
)
|
190 |
+
|
191 |
+
# MLM auxiliary objective
|
192 |
+
if mlm_input_ids is not None:
|
193 |
+
mlm_outputs = self.roberta(
|
194 |
+
input_ids=mlm_input_ids,
|
195 |
+
attention_mask=mlm_attention_mask,
|
196 |
+
token_type_ids=None,
|
197 |
+
position_ids=None,
|
198 |
+
head_mask=None,
|
199 |
+
inputs_embeds=None,
|
200 |
+
output_attentions=output_attentions,
|
201 |
+
output_hidden_states=output_hidden_states,
|
202 |
+
return_dict=return_dict,
|
203 |
+
)
|
204 |
+
|
205 |
+
# Pooling
|
206 |
+
user_pooler_output = self.pooler(user_attention_mask, user_outputs)
|
207 |
+
pos_item_pooler_output = self.pooler(pos_item_attention_mask, pos_item_outputs)
|
208 |
+
neg_item_pooler_output = self.pooler(neg_item_attention_mask, neg_item_outputs)
|
209 |
+
|
210 |
+
# If using "cls", we add an extra MLP layer
|
211 |
+
# (same as BERT's original implementation) over the representation.
|
212 |
+
if self.pooler_type == "cls":
|
213 |
+
user_pooler_output = self.mlp(user_pooler_output)
|
214 |
+
pos_item_pooler_output = self.mlp(pos_item_pooler_output)
|
215 |
+
neg_item_pooler_output = self.mlp(neg_item_pooler_output)
|
216 |
+
|
217 |
+
# Gather all item embeddings if using distributed training
|
218 |
+
if dist.is_initialized() and self.training:
|
219 |
+
# Dummy vectors for allgather
|
220 |
+
user_list = [torch.zeros_like(user_pooler_output) for _ in range(dist.get_world_size())]
|
221 |
+
pos_item_list = [torch.zeros_like(pos_item_pooler_output) for _ in range(dist.get_world_size())]
|
222 |
+
neg_item_list = [torch.zeros_like(neg_item_pooler_output) for _ in range(dist.get_world_size())]
|
223 |
+
# Allgather
|
224 |
+
dist.all_gather(tensor_list=user_list, tensor=user_pooler_output.contiguous())
|
225 |
+
dist.all_gather(tensor_list=pos_item_list, tensor=pos_item_pooler_output.contiguous())
|
226 |
+
dist.all_gather(tensor_list=neg_item_list, tensor=neg_item_pooler_output.contiguous())
|
227 |
+
|
228 |
+
# Since allgather results do not have gradients, we replace the
|
229 |
+
# current process's corresponding embeddings with original tensors
|
230 |
+
user_list[dist.get_rank()] = user_pooler_output
|
231 |
+
pos_item_list[dist.get_rank()] = pos_item_pooler_output
|
232 |
+
neg_item_list[dist.get_rank()] = neg_item_pooler_output
|
233 |
+
|
234 |
+
# Get full batch embeddings
|
235 |
+
user_pooler_output = torch.cat(user_list, dim=0)
|
236 |
+
pos_item_pooler_output = torch.cat(pos_item_list, dim=0)
|
237 |
+
neg_item_pooler_output = torch.cat(neg_item_list, dim=0)
|
238 |
+
|
239 |
+
cos_sim = self.sim(user_pooler_output.unsqueeze(1), pos_item_pooler_output.unsqueeze(0))
|
240 |
+
neg_sim = self.sim(user_pooler_output.unsqueeze(1), neg_item_pooler_output.unsqueeze(0))
|
241 |
+
cos_sim = torch.cat([cos_sim, neg_sim], 1)
|
242 |
+
|
243 |
+
labels = torch.arange(cos_sim.size(0)).long().to(self.device)
|
244 |
+
loss_fct = nn.CrossEntropyLoss()
|
245 |
+
|
246 |
+
loss = loss_fct(cos_sim, labels)
|
247 |
+
|
248 |
+
# Calculate loss for MLM
|
249 |
+
if mlm_outputs is not None and mlm_labels is not None and self.model_args.do_mlm:
|
250 |
+
mlm_labels = mlm_labels.view(-1, mlm_labels.size(-1))
|
251 |
+
prediction_scores = self.lm_head(mlm_outputs.last_hidden_state)
|
252 |
+
masked_lm_loss = loss_fct(prediction_scores.view(-1, self.config.vocab_size), mlm_labels.view(-1))
|
253 |
+
loss = loss + self.model_args.mlm_weight * masked_lm_loss
|
254 |
+
|
255 |
+
if not return_dict:
|
256 |
+
raise NotImplementedError
|
257 |
+
|
258 |
+
return SequenceClassifierOutput(
|
259 |
+
loss=loss,
|
260 |
+
logits=cos_sim,
|
261 |
+
)
|
262 |
+
|
263 |
+
def encode(self,
|
264 |
+
input_ids=None,
|
265 |
+
attention_mask=None,
|
266 |
+
token_type_ids=None,
|
267 |
+
position_ids=None,
|
268 |
+
head_mask=None,
|
269 |
+
inputs_embeds=None,
|
270 |
+
labels=None,
|
271 |
+
output_attentions=None,
|
272 |
+
output_hidden_states=None,
|
273 |
+
return_dict=None,
|
274 |
+
):
|
275 |
+
return_dict = return_dict if return_dict is not None else self.config.use_return_dict
|
276 |
+
outputs = self.roberta(
|
277 |
+
input_ids=input_ids,
|
278 |
+
attention_mask=attention_mask,
|
279 |
+
token_type_ids=None,
|
280 |
+
position_ids=None,
|
281 |
+
head_mask=None,
|
282 |
+
inputs_embeds=None,
|
283 |
+
output_attentions=output_attentions,
|
284 |
+
output_hidden_states=output_hidden_states,
|
285 |
+
return_dict=return_dict,
|
286 |
+
)
|
287 |
+
pooler_output = self.pooler(attention_mask, outputs)
|
288 |
+
if self.pooler_type == "cls":
|
289 |
+
pooler_output = self.mlp(pooler_output)
|
290 |
+
if not return_dict:
|
291 |
+
return (outputs[0], pooler_output) + outputs[2:]
|
292 |
+
|
293 |
+
return BaseModelOutputWithPoolingAndCrossAttentions(
|
294 |
+
pooler_output=pooler_output,
|
295 |
+
last_hidden_state=outputs.last_hidden_state,
|
296 |
+
hidden_states=outputs.hidden_states,
|
297 |
+
)
|
298 |
+
|
299 |
+
def inference(self,
|
300 |
+
user_profile_list,
|
301 |
+
item_profile_list,
|
302 |
+
dataset_name,
|
303 |
+
tokenizer,
|
304 |
+
infer_batch_size=128
|
305 |
+
):
|
306 |
+
n_user = len(user_profile_list)
|
307 |
+
profiles = user_profile_list + item_profile_list
|
308 |
+
n_batch = math.ceil(len(profiles) / infer_batch_size)
|
309 |
+
text_embeds = []
|
310 |
+
for i in tqdm(range(n_batch), desc=f'Encoding Text {dataset_name}'):
|
311 |
+
batch_profiles = profiles[i * infer_batch_size: (i + 1) * infer_batch_size]
|
312 |
+
inputs = tokenizer(batch_profiles, padding=True, truncation=True, max_length=512, return_tensors="pt")
|
313 |
+
for k in inputs:
|
314 |
+
inputs[k] = inputs[k].to(self.device)
|
315 |
+
with torch.inference_mode():
|
316 |
+
embeds = self.encode(
|
317 |
+
input_ids=inputs.input_ids,
|
318 |
+
attention_mask=inputs.attention_mask
|
319 |
+
)
|
320 |
+
text_embeds.append(embeds.pooler_output.detach().cpu())
|
321 |
+
text_embeds = torch.concat(text_embeds, dim=0).cuda()
|
322 |
+
user_embeds = F.normalize(text_embeds[: n_user], dim=-1)
|
323 |
+
item_embeds = F.normalize(text_embeds[n_user: ], dim=-1)
|
324 |
+
return user_embeds, item_embeds
|
|