idea.py

# automatic pipeline
pipe=Pipeline('summarization',model="bart")
pipe()
# preprocess(input)
# model(input)
# postprocess(input)


#-------------------------------------------------------------------------#

# manual pipeline
model_id='bart-large-cnn'
tokenizer='bart-large-cnn'

preprocess_summary  = PreProcessSummaryPipeline(tokenizer)
model_summary       = PytorchModelPipeline(tokenizer)
postprocess_summary = PostProcessSummaryPipeline(tokenizer)

pre  = preprocess_summary(input)
pred = ModelSummaryPipeline(pre)
post = postprocess_summary(pred)

#-------------------------------------------------------------------------#

# chained pipeline

# summary steps
model_id='bart-large-cnn'
tokenizer='bart-large-cnn'

preprocess_summary  = PreProcessSummaryPipeline(tokenizer)
model_summary       = PytorchModelPipeline(model_id)
postprocess_summary = PostProcessSummaryPipeline(tokenizer)

# token classification steps
model_id='my-onnx-model'
tokenizer='bert-base-cased'

preprocess_token    = PreProcessSummaryPipeline(tokenizer)
model_token         = ONNXModelPipeline(model_id)
postprocess_model   = PostProcessSummaryPipeline(tokenizer)

pipeline=Pipeline([
		preproces_summary,
		model_summary,
		postprocess_summary,
		preprocess_token,
		model_token,
		postprocess_model
])
pipeline(inuput)
# preprocess_summary(input)
# model_summary(input)
# postprocess_summary(input)
# preprocess_token(input)
# model_token(input)
# return postprocess_model(input)


#-------------------------------------------------------------------------#

# remote pipeline
preprocess_summary   = PreProcessSummaryPipeline()
remote_model_summary = TritonClientPipeline() # wrapper for e.g. triton.client with __call__
postprocess_summary  = PostProcessSummaryPipeline()

pipeline=Pipeline([preproces_summary,remote_model_summary,postprocess_summary])
pipeline()

#-------------------------------------------------------------------------#

# parallel pipeline
preprocess       = PreProcessPipeline()
model_sentiment  = PytorchModelPipeline()
model_news_class = PytorchModelPipeline()
postprocess      = PostProcessPipeline()

pipeline=Pipeline([preprocess,[model_sentiment,model_news_class],postprocess])
pipeline()