jinyong-gpt2

飞雪连天射白鹿，笑书神侠倚碧鸳

Model description

AI生成金庸小说，给出开头续写。

How to use

使用 pipeline 调用模型:

>>> # 调用微调后的模型
>>> senc="这些雪花落下来,多么白,多么好看.过几天太阳出来,每一片 雪花都变得无影无踪.到得明年冬天,又有许很多多雪花,只不过已不是 今年这些雪花罢了。"
>>> model_id="jinyong-gpt2-finetuning"
>>> from transformers import AutoTokenizer, GPT2LMHeadModel, TextGenerationPipeline

>>> tokenizer = AutoTokenizer.from_pretrained(model_id) 
>>> model = GPT2LMHeadModel.from_pretrained(model_id)
>>> text_generator = TextGenerationPipeline(model, tokenizer)   
>>> text_generator.model.config.pad_token_id = text_generator.model.config.eos_token_id
>>> text_generator( senc,max_length=108, do_sample=True)
[{'generated_text': '这些雪花落下来,多么白,多么好看.过几天太阳出来,每一片 雪花都变得无影无踪.到得明年冬天,又有许很多多雪花,只不过已不是 今年这些雪花罢了。 反正 老天爷 有眼 ， 不知 哪里 是甚么 风 险 ？” 正 说到此处 ， 突然 听得 谢逊 啸声 渐近 ， 忍不住 张口 惊呼 ， 一齐 向他 扑去 ， 只听 谢逊 一声 怒吼 ， 跟着 左手 用力 拍 出一掌 ， 以 掌力 化开 。 众人 吃了一惊 ， 同时 从 海 道 中 跃出 ， 双双 倒退 。 张翠山和殷素素 对望一眼 ， 均想 以 这两 大高手 之力 如何 抵挡 ， 以 今日 之力 如何 攻敌 之'}]
>>> 

Here is how to use this model to get the features of a given text in PyTorch:

from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("supermy/jinyong-gpt2")

model = AutoModelForCausalLM.from_pretrained("supermy/jinyong-gpt2")

Training data

此数据集基于金庸的【飞雪连天射白鹿，笑书神侠倚碧鸳】小说集训练。

统计信息

Training procedure

基于模型：GPT2 训练环境：英伟达16G显卡

bpe分词："vocab_size"=30000

[INFO|trainer.py:1608] 2022-12-02 19:52:59,024 >> ***** Running training *****
[INFO|trainer.py:1609] 2022-12-02 19:52:59,024 >>   Num examples = 9443
[INFO|trainer.py:1610] 2022-12-02 19:52:59,024 >>   Num Epochs = 108
[INFO|trainer.py:1611] 2022-12-02 19:52:59,024 >>   Instantaneous batch size per device = 12
[INFO|trainer.py:1612] 2022-12-02 19:52:59,024 >>   Total train batch size (w. parallel, distributed & accumulation) = 12
[INFO|trainer.py:1613] 2022-12-02 19:52:59,024 >>   Gradient Accumulation steps = 1
[INFO|trainer.py:1614] 2022-12-02 19:52:59,024 >>   Total optimization steps = 84996
[INFO|trainer.py:1616] 2022-12-02 19:52:59,025 >>   Number of trainable parameters = 124439808

[INFO|trainer.py:1608] 2022-12-03 21:44:00,182 >> ***** Running training *****
[INFO|trainer.py:1609] 2022-12-03 21:44:00,182 >>   Num examples = 9443
[INFO|trainer.py:1610] 2022-12-03 21:44:00,182 >>   Num Epochs = 216
[INFO|trainer.py:1611] 2022-12-03 21:44:00,182 >>   Instantaneous batch size per device = 12
[INFO|trainer.py:1612] 2022-12-03 21:44:00,182 >>   Total train batch size (w. parallel, distributed & accumulation) = 12
[INFO|trainer.py:1613] 2022-12-03 21:44:00,182 >>   Gradient Accumulation steps = 1
[INFO|trainer.py:1614] 2022-12-03 21:44:00,182 >>   Total optimization steps = 169992
[INFO|trainer.py:1616] 2022-12-03 21:44:00,183 >>   Number of trainable parameters = 124439808
[INFO|trainer.py:1637] 2022-12-03 21:44:00,184 >>   Continuing training from checkpoint, will skip to saved global_step
[INFO|trainer.py:1638] 2022-12-03 21:44:00,184 >>   Continuing training from epoch 107
[INFO|trainer.py:1639] 2022-12-03 21:44:00,184 >>   Continuing training from global step 84500

[INFO|trainer.py:1608] 2022-12-05 07:36:13,626 >> ***** Running training *****
[INFO|trainer.py:1609] 2022-12-05 07:36:13,626 >>   Num examples = 9443
[INFO|trainer.py:1610] 2022-12-05 07:36:13,626 >>   Num Epochs = 368
[INFO|trainer.py:1611] 2022-12-05 07:36:13,626 >>   Instantaneous batch size per device = 12
[INFO|trainer.py:1612] 2022-12-05 07:36:13,626 >>   Total train batch size (w. parallel, distributed & accumulation) = 12
[INFO|trainer.py:1613] 2022-12-05 07:36:13,626 >>   Gradient Accumulation steps = 1
[INFO|trainer.py:1614] 2022-12-05 07:36:13,626 >>   Total optimization steps = 289616
[INFO|trainer.py:1616] 2022-12-05 07:36:13,627 >>   Number of trainable parameters = 124439808
[INFO|trainer.py:1637] 2022-12-05 07:36:13,628 >>   Continuing training from checkpoint, will skip to saved global_step
[INFO|trainer.py:1638] 2022-12-05 07:36:13,628 >>   Continuing training from epoch 255
[INFO|trainer.py:1639] 2022-12-05 07:36:13,628 >>   Continuing training from global step 201000

{'loss': 8.0431, 'learning_rate': 4.970998635229893e-05, 'epoch': 0.64}
{'loss': 7.4867, 'learning_rate': 4.94158548637583e-05, 'epoch': 1.27}
{'loss': 7.322, 'learning_rate': 4.912172337521766e-05, 'epoch': 1.91}
......
{'loss': 3.901, 'learning_rate': 2.5010882865076008e-05, 'epoch': 108.01}
{'loss': 3.8959, 'learning_rate': 2.4863817120805686e-05, 'epoch': 108.64}
......
{'loss': 3.1625, 'learning_rate': 4.6090404254317857e-07, 'epoch': 214.1}
{'loss': 3.1592, 'learning_rate': 3.1413242976140055e-07, 'epoch': 214.74}
{'loss': 3.1625, 'learning_rate': 1.6706668549108195e-07, 'epoch': 215.37}
{'train_runtime': 72271.9602, 'train_samples_per_second': 28.222, 'train_steps_per_second': 2.352, 'train_loss': 1.7180436183842016, 'epoch': 216.0}
{'loss': 2.7087, 'learning_rate': 4.2642671675598036e-08, 'epoch': 367.85}
{'train_runtime': 74859.0808, 'train_samples_per_second': 46.421, 'train_steps_per_second': 3.869, 'train_loss': 0.8725239146935282, 'epoch': 368.0}
***** train metrics *****
  epoch                    =       368.0
  train_loss               =      0.8725
  train_runtime            = 20:47:39.08
  train_samples            =        9443
  train_samples_per_second =      46.421
  train_steps_per_second   =       3.869
12/06/2022 04:23:55 - INFO - __main__ - *** Evaluate ***
[INFO|trainer.py:2929] 2022-12-06 04:23:55,953 >> ***** Running Evaluation *****
[INFO|trainer.py:2931] 2022-12-06 04:23:55,953 >>   Num examples = 283
[INFO|trainer.py:2934] 2022-12-06 04:23:55,954 >>   Batch size = 12
100%|██████████| 24/24 [00:07<00:00,  3.20it/s]
[INFO|modelcard.py:449] 2022-12-06 04:24:04,760 >> Dropping the following result as it does not have all the necessary fields:
{'task': {'name': 'Causal Language Modeling', 'type': 'text-generation'}, 'metrics': [{'name': 'Accuracy', 'type': 'accuracy', 'value': 0.19599206157122803}]}
***** eval metrics *****
  epoch                   =      368.0
  eval_accuracy           =      0.196
  eval_loss               =     7.9524
  eval_runtime            = 0:00:07.87
  eval_samples            =        283
  eval_samples_per_second =      35.94
  eval_steps_per_second   =      3.048
  perplexity              =  2842.2766