MCPcopy Create free account
hub / github.com/NVIDIA/TensorRT-LLM / prepare_text_inputs

Function prepare_text_inputs

examples/models/core/bert/utils.py:40–76  ·  view source on GitHub ↗
(model_name, batch_size=8)

Source from the content-addressed store, hash-verified

38
39
40def prepare_text_inputs(model_name, batch_size=8):
41 print(
42 f"HF_DATASETS_OFFLINE inside function: {datasets.config.HF_DATASETS_OFFLINE}"
43 )
44 if model_name == "BertForQuestionAnswering" or model_name == "RobertaForQuestionAnswering":
45 squad_dataset_root = str(llm_datasets_root(
46 )) + "/" if datasets.config.HF_DATASETS_OFFLINE else ""
47 squad_dataset_path = squad_dataset_root + "squad_v2"
48 squad_dataset = load_dataset(squad_dataset_path, trust_remote_code=True)
49 val_dataset = squad_dataset["validation"]
50 samples = val_dataset.select(range(batch_size))
51
52 qa_real_test_inputs = {
53 'text': samples["question"],
54 'text_pair': samples["context"]
55 }
56 return qa_real_test_inputs
57 elif model_name == "BertForSequenceClassification" or model_name == "RobertaForSequenceClassification":
58 yelp_dataset_root = str(llm_datasets_root(
59 )) + "/" if datasets.config.HF_DATASETS_OFFLINE else "fancyzhx/"
60 yelp_dataset_path = yelp_dataset_root + "yelp_polarity"
61 yelp_dataset = load_dataset(yelp_dataset_path, trust_remote_code=True)
62 val_dataset = yelp_dataset["test"]
63 samples = val_dataset.select(range(batch_size))
64
65 seqcls_real_test_inputs = {'text': samples['text']}
66 return seqcls_real_test_inputs
67 elif model_name == "BertModel" or model_name == "RobertaModel":
68 #NOTE: For BertModel, it is used as an encoder, so we use dummy input here,
69 # you can choose whatevert you like, but the numerical accuracy might vary.
70 test_input = 'To be or not to be: that is the question'
71 input_strings = [test_input for _ in range(batch_size)]
72 base_real_test_inputs = {'text': input_strings}
73 return base_real_test_inputs
74
75 else:
76 raise NotImplementedError(f"Unknown model {model_name}")
77
78
79def get_engine_name(rank):

Callers 1

run.pyFile · 0.90

Calls 3

selectMethod · 0.80
llm_datasets_rootFunction · 0.70
load_datasetFunction · 0.50

Tested by

no test coverage detected