python打造个性化ai_用迁移学习打造最先进会话AI-CSDN博客

本文链接：https://blog.csdn.net/weixin_28694295/article/details/112032723

这个代码库提供了一个干净的、注释过的代码，用于训练和测试对话代理，利用OpenAI GPT和GPT-2语言模型的迁移学习。通过在8个V100云实例上训练大约一小时，可以实现最先进的会话AI。代码还提供了预训练模型，可直接下载使用。

摘要由CSDN通过智能技术生成

🦄 Building a State-of-the-Art Conversational AI with Transfer Learning

This code is a clean and commented code base with training and testing scripts that can be used to train a dialog agent leveraging transfer Learning from an OpenAI GPT and GPT-2 Transformer language model.

This codebase can be used to reproduce the results of HuggingFace's participation to NeurIPS 2018 dialog competition ConvAI2 which was state-of-the-art on the automatic metrics. The 3k+ lines of competition code was distilled in about 250 lines of training code with distributed & FP16 options to form the present repository.

This model can be trained in about one hour on a 8 V100 cloud instance (currently costs about $25) and a pre-trained model is also made available.

Installation

To install and use the training and inference scripts please clone the repo and install the requirements:

git clone https://github.com/huggingface/transfer-learning-conv-ai

cd transfer-learning-conv-ai

pip install -r requirements.txt

python -m spacy download en

Installation with Docker

To install using docker please build the self-contained image:

docker build -t convai .

You can then enter the image

ip-192-168-22-157:transfer-learning-conv-ai loretoparisi$ docker run --rm -it convai bash

root@91e241bb823e:/# ls

Dockerfile README.md boot dev home lib media models proc root sbin sys train.py utils.py

LICENCE bin convai_evaluation.py etc interact.py lib64 mnt opt requirements.txt run srv tmp usr var

You can then run the interact.py script on the pretrained model:

python3 interact.py --model models/

Pretrained model

We make a pretrained and fine-tuned model available on our S3 here. The easiest way to download and use this model is just to run the interact.py script to talk with the model. Without any argument, this script will automatically download and cache our model.

Using the training script

The training script can be used in single GPU or multi GPU settings:

python ./train.py # Single GPU training

python -m torch.distributed.launch --nproc_per_node=8 ./train.py # Training on 8 GPUs

The training script accept several arguments to tweak the training:

Argument

Type

Default value

Description

dataset_path

str

Path or url of the dataset. If empty download from S3.

dataset_cache

str

'./dataset_cache.bin'

Path or url of the dataset cache

model

str

"openai-gpt"

Path, url or short name of the model

num_candidates

int

Number of candidates for training

max_history

int

Number of previous exchanges to keep in history

train_batch_size

int

Batch size for training

valid_batch_size

int

Batch size for validation

gradient_accumulation_steps

int

Accumulate gradients on several steps

float

6.25e-5

Learning rate

lm_coef

float

1.0

LM loss coefficient

mc_coef

float

1.0

Multiple-choice loss coefficient

max_norm

float

1.0

Clipping gradient norm

n_epochs

int

Number of training epochs

personality_permutations

int

Number of permutations of personality sentences

device

str

"cuda" if torch.cuda.is_available() else "cpu"

Device (cuda or cpu)

fp16

str

Set to O0, O1, O2 or O3 for fp16 training (see apex documentation)

local_rank

int

-1

Local rank for distributed training (-1: not distributed)

Here is how to reproduce our results on a server with 8 V100 GPUs (adapt number of nodes and batch sizes to your configuration):

python -m torch.distributed.launch --nproc_per_node=8 ./train.py --gradient_accumulation_steps=4 --lm_coef=2.0 --max_history=2 --n_epochs=1 --num_candidates=4 --personality_permutations=2 --train_batch_size=2 --valid_batch_size=2

This model should give a Hits@1 over 79, perplexity of 20.5 and F1 of 16.5 using the convai2 evaluation script (see below).

These numbers are slightly lower than the number we obtained in the ConvAI2 competition. Here is what you can tweak to reach the same results:

in the ConvAI2 competition we also used tweaked position emebddings so that the history of the dialog always start at with the same embeddings. This is easy to add with pytorch-transformers and should improve the hits@1 metric.

in the ConvAI2 competition we used a beam search decoder. While the results are better in term of f1 metric, our feeling is that the human experience is less compelling with beam search versus the nucleus sampling detector which is provided in the present repository.

Using the interaction script

The training script saves all the experiments and checkpoints in a sub-folder named with the timestamp of the experiment in the ./runs folder of the repository base folder.

You can then use the interactive script to interact with the model simply by pointing to this folder.

Here is an example command line to run the interactive script:

python ./interact.py --model_checkpoint ./data/Apr17_13-31-38_thunder/ # run the interactive script with a training checkpoint

python ./interact.py # run the interactive script with the finetuned model on our S3

The fine-tuned model will gives FINAL Hits@1: 0.715

The interactive script accept a few arguments to tweak the decoding algorithm:

Argument

Type

Default value

Description

dataset_path

str

Path or url of the dataset. If empty download from S3.

dataset_cache

str

'./dataset_cache.bin'

Path or url of the dataset cache

model

str

"openai-gpt"

Path, url or short name of the model

max_history

int

Number of previous utterances to keep in history

device

str

cuda if torch.cuda.is_available() else cpu

Device (cuda or cpu)

no_sample

action store_true

Set to use greedy decoding instead of sampling

max_length

int

Maximum length of the output utterances

min_length

int

Minimum length of the output utterances

seed

int

Seed

temperature

int

0.7

Sampling softmax temperature

top_k

int

Filter top-k tokens before sampling (<=0: no filtering)

top_p

float

0.9

Nucleus filtering (top-p) before sampling (<=0.0: no filtering)

Running ConvAI2 evaluation scripts

To run the evaluation scripts of the ConvAI2 challenge, you first need to install ParlAI in the repo base folder like this:

git clone https://github.com/facebookresearch/ParlAI.git

cd ParlAI

python setup.py develop

You can then run the evaluation script from ParlAI base folder:

cd ParlAI

python ../convai_evaluation.py --eval_type hits@1 # to download and evaluate our fine-tuned model on hits@1 metric

python ../convai_evaluation.py --eval_type hits@1 --model_checkpoint ./data/Apr17_13-31-38_thunder/ # to evaluate a training checkpoint on hits@1 metric

The evaluation script accept a few arguments to select the evaluation metric and tweak the decoding algorithm:

Argument

Type

Default value

Description

eval_type

str

"hits@1"

Evaluate the model on hits@1, ppl or f1 metric on the ConvAI2 validation dataset

model

str

"openai-gpt"