I am using H2O DAI 1.9.0.6. I am tring to load custom recipe (BERT pretained model using custom recipe) on Expert settings. I am using local file to upload. However upload is not happning. No error, no progress nothing. After that activity I am not able to see this model under RECIPE tab.
Took Sample Recipe from below URL and Modified for my need. Thanks for the person who created this Recipe.
https://github.com/h2oai/driverlessai-recipes/blob/master/models/nlp/portuguese_bert.py
Custom Recipe
import os
import shutil
from urllib.parse import urlparse
import requests
from h2oaicore.models import TextBERTModel, CustomModel
from h2oaicore.systemutils import make_experiment_logger, temporary_files_path, atomic_move, loggerinfo
def is_url(url):
try:
result = urlparse(url)
return all([result.scheme, result.netloc, result.path])
except:
return False
def maybe_download_language_model(logger,
save_directory,
model_link,
config_link,
vocab_link):
model_name = "pytorch_model.bin"
if isinstance(model_link, str):
model_name = model_link.split('/')[-1]
if '.bin' not in model_name:
model_name = "pytorch_model.bin"
maybe_download(url=config_link,
dest=os.path.join(save_directory, "config.json"),
logger=logger)
maybe_download(url=vocab_link,
dest=os.path.join(save_directory, "vocab.txt"),
logger=logger)
maybe_download(url=model_link,
dest=os.path.join(save_directory, model_name),
logger=logger)
def maybe_download(url, dest, logger=None):
if not is_url(url):
loggerinfo(logger, f"{url} is not a valid URL.")
return
dest_tmp = dest + ".tmp"
if os.path.exists(dest):
loggerinfo(logger, f"already downloaded {url} -> {dest}")
return
if os.path.exists(dest_tmp):
loggerinfo(logger, f"Download has already started {url} -> {dest_tmp}. "
f"Delete {dest_tmp} to download the file once more.")
return
loggerinfo(logger, f"Downloading {url} -> {dest}")
url_data = requests.get(url, stream=True)
if url_data.status_code != requests.codes.ok:
msg = "Cannot get url %s, code: %s, reason: %s" % (
str(url), str(url_data.status_code), str(url_data.reason))
raise requests.exceptions.RequestException(msg)
url_data.raw.decode_content = True
if not os.path.isdir(os.path.dirname(dest)):
os.makedirs(os.path.dirname(dest), exist_ok=True)
with open(dest_tmp, 'wb') as f:
shutil.copyfileobj(url_data.raw, f)
atomic_move(dest_tmp, dest)
def check_correct_name(custom_name):
allowed_pretrained_models = ['bert', 'openai-gpt', 'gpt2', 'transfo-xl', 'xlnet', 'xlm-roberta',
'xlm', 'roberta', 'distilbert', 'camembert', 'ctrl', 'albert']
assert len([model_name for model_name in allowed_pretrained_models
if model_name in custom_name]), f"{custom_name} needs to contain the name" \
" of the pretrained model architecture (e.g. bert or xlnet) " \
"to be able to process the model correctly."
class CustomBertModel(TextBERTModel, CustomModel):
"""
Custom model class for using pretrained transformer models.
The class inherits :
- CustomModel that really is just a tag. It's there to make sure DAI knows it's a custom model.
- TextBERTModel so that the custom model inherits all the properties and methods.
Supported model architecture:
'bert', 'openai-gpt', 'gpt2', 'transfo-xl', 'xlnet', 'xlm-roberta',
'xlm', 'roberta', 'distilbert', 'camembert', 'ctrl', 'albert'
How to use:
- You have already downloaded the weights, the vocab and the config file:
- Set _model_path as the folder where the weights, the vocab and the config file are stored.
- Set _model_name according to the pretrained architecture (e.g. bert-base-uncased).
- You want to to download the weights, the vocab and the config file:
- Set _model_link, _config_link and _vocab_link accordingly.
- _model_path is the folder where the weights, the vocab and the config file will be saved.
- Set _model_name according to the pretrained architecture (e.g. bert-base-uncased).
- Important:
_model_path needs to contain the name of the pretrained model architecture (e.g. bert or xlnet)
to be able to load the model correctly.
- Disable genetic algorithm in the expert setting.
"""
# _model_path is the full path to the directory where the weights, vocab and the config will be saved.
_model_name = NotImplemented # Will be used to create the MOJO
_model_path = NotImplemented
_model_link = NotImplemented
_config_link = NotImplemented
_vocab_link = NotImplemented
_booster_str = "pytorch-custom"
# Requirements for MOJO creation:
# _model_name needs to be one of
# bert-base-uncased, bert-base-multilingual-cased, xlnet-base-cased, roberta-base, distilbert-base-uncased
# vocab.txt needs to be the same as vocab.txt used in _model_name (no custom vocabulary yet).
_mojo = False
@staticmethod
def is_enabled():
return False # Abstract Base model should not show up in models.
def _set_model_name(self, language_detected):
self.model_path = self.__class__._model_path
self.model_name = self.__class__._model_name
check_correct_name(self.model_path)
check_correct_name(self.model_name)
def fit(self, X, y, sample_weight=None, eval_set=None, sample_weight_eval_set=None, **kwargs):
logger = None
if self.context and self.context.experiment_id:
logger = make_experiment_logger(experiment_id=self.context.experiment_id, tmp_dir=self.context.tmp_dir,
experiment_tmp_dir=self.context.experiment_tmp_dir)
maybe_download_language_model(logger,
save_directory=self.__class__._model_path,
model_link=self.__class__._model_link,
config_link=self.__class__._config_link,
vocab_link=self.__class__._vocab_link)
super().fit(X, y, sample_weight, eval_set, sample_weight_eval_set, **kwargs)
class GermanBertModel(CustomBertModel):
_model_name = "bert-base-german-dbmdz-uncased"
_model_path = os.path.join(temporary_files_path, "german_bert_language_model/")
_model_link = "https://huggingface.co/bert-base-german-dbmdz-uncased/resolve/main/pytorch_model.bin"
_config_link = "https://huggingface.co/bert-base-german-dbmdz-uncased/resolve/main/config.json"
_vocab_link = "https://huggingface.co/bert-base-german-dbmdz-uncased/resolve/main/vocab.txt"
_mojo = True
@staticmethod
def is_enabled():
return True
Check that your custom recipe has is_enabled() returning True.
def is_enabled():
return True