Spaces:
Runtime error
Runtime error
File size: 2,625 Bytes
7a1c034 2059531 7a1c034 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 |
import torch
import numpy as np
from transformers import AutoTokenizer, AutoModelForSeq2SeqLM
def getScores(ids, scores, pad_token_id):
"""get sequence scores from model.generate output"""
scores = torch.stack(scores, dim=1)
log_probs = torch.log_softmax(scores, dim=2)
# remove start token
ids = ids[:,1:]
# gather needed probs
x = ids.unsqueeze(-1).expand(log_probs.shape)
needed_logits = torch.gather(log_probs, 2, x)
final_logits = needed_logits[:, :, 0]
padded_mask = (ids == pad_token_id)
final_logits[padded_mask] = 0
final_scores = final_logits.sum(dim=-1)
return final_scores.cpu().detach().numpy()
def topkSample(input, model, tokenizer,
num_samples=5,
num_beams=1,
max_output_length=30):
tokenized = tokenizer(input, return_tensors="pt")
out = model.generate(**tokenized,
do_sample=True,
num_return_sequences = num_samples,
num_beams = num_beams,
eos_token_id = tokenizer.eos_token_id,
pad_token_id = tokenizer.pad_token_id,
output_scores = True,
return_dict_in_generate=True,
max_length=max_output_length,)
out_tokens = out.sequences
out_str = tokenizer.batch_decode(out_tokens, skip_special_tokens=True)
out_scores = getScores(out_tokens, out.scores, tokenizer.pad_token_id)
pair_list = [(x[0], x[1]) for x in zip(out_str, out_scores)]
sorted_pair_list = sorted(pair_list, key=lambda x:x[1], reverse=True)
return sorted_pair_list
def greedyPredict(input, model, tokenizer):
input_ids = tokenizer([input], return_tensors="pt").input_ids
out_tokens = model.generate(input_ids)
out_str = tokenizer.batch_decode(out_tokens, skip_special_tokens=True)
return out_str[0]
def predict_tail(entity, relation):
global model, tokenizer
input = entity + "| " + relation
out = topkSample(input, model, tokenizer, num_samples=5)
out_dict = {}
for k, v in out:
out_dict[k] = np.exp(v).item()
return out_dict
tokenizer = AutoTokenizer.from_pretrained("apoorvumang/kgt5-wikikg90mv2")
model = AutoModelForSeq2SeqLM.from_pretrained("apoorvumang/kgt5-base-wikikg90mv2")
ent_input = gradio.inputs.Textbox(lines=1, default="World War II")
rel_input = gradio.inputs.Textbox(lines=1, default="followed by")
output = gradio.outputs.Label()
iface = gr.Interface(fn=predict_tail, inputs=[ent_input, rel_input], outputs=output)
iface.launch()
|