政务问答系统模型动转静,插入milvus

def gen_id2corpus(corpus_file):
    id2corpus = {}
    with open(corpus_file, "r", encoding="utf-8") as f:
        for idx, line in enumerate(f):
            id2corpus[idx] = line.rstrip()
    return id2corpus

corpus_file='./datasets/zwqa/corpus.csv'
id2corpus = gen_id2corpus(corpus_file)
corpus_list = [{idx: text} for idx, text in id2corpus.items()]

model_dir='./output'
model_file = model_dir + "/inference.get_pooled_embedding.pdmodel"
params_file = model_dir + "/inference.get_pooled_embedding.pdiparams"

import os
import paddle

if not os.path.exists(model_file):
    raise ValueError("not find model file path {}".format(model_file))
if not os.path.exists(params_file):
    raise ValueError("not find params file path {}".format(params_file))
config = paddle.inference.Config(model_file, params_file)

from paddle import inference

precision='fp32'
output_emb_size = 256

config.enable_use_gpu(100, 0)
precision_map = {
            "fp16": inference.PrecisionType.Half,
            "fp32": inference.PrecisionType.Float32,
            "int8": inference.PrecisionType.Int8,
        }
precision_mode = precision_map[precision]

batch_size=32

# config.enable_tensorrt_engine(max_batch_size=batch_size,
#                                               min_subgraph_size=30,
#                                               precision_mode=precision_mode)

 predictor.get_input_names()

config.switch_use_feed_fetch_ops(False)
predictor = paddle.inference.create_predictor(config)
input_handles = [predictor.get_input_handle(name) for name in predictor.get_input_names()]
output_handle = predictor.get_output_handle(predictor.get_output_names()[0])

from paddlenlp.data import Pad, Tuple

from paddlenlp.transformers import AutoTokenizer

tokenizer = AutoTokenizer.from_pretrained('ernie-3.0-medium-zh')

def batchify_fn(
            samples,
            fn=Tuple(
                Pad(axis=0, pad_val=tokenizer.pad_token_id, dtype="int64"),  # input
                Pad(axis=0, pad_val=tokenizer.pad_token_id, dtype="int64"),  # segment
            ),
        ):
            return fn(samples)

corpus_list[:3]#索引文本集，与问答对中的问题相对应，用于构建标准问题语义库

def convert_example(example, tokenizer, max_seq_length=512