text_generation uses the TextGenerationInput schema which has an alias prompt for sequences. This PR updates the data_creation function such that benchmarking text_generation works. Was previously failing due to incorrect schema creation
Testing
Testing locally using an mpt model:
deepsparse.benchmark_pipeline text_generation zoo:mpt-7b-mpt_pretrain-base_quantized -x v1.json
Was previously getting this error:
pydantic.error_wrappers.ValidationError: 1 validation error for TextGenerationInput
prompt
field required (type=value_error.missing)
Summary
text_generation
uses theTextGenerationInput
schema which has an aliasprompt
for sequences. This PR updates thedata_creation
function such that benchmarking text_generation works. Was previously failing due to incorrect schema creationTesting
deepsparse.benchmark_pipeline text_generation zoo:mpt-7b-mpt_pretrain-base_quantized -x v1.json
Was previously getting this error: