MCPcopy Index your code
hub / github.com/THUDM/GLM / build_sample

Function build_sample

tasks/data_utils.py:266–298  ·  view source on GitHub ↗

Convert to numpy and return a sample consumed by the batch producer.

(ids, types=None, paddings=None, positions=None, masks=None, label=None, unique_id=None, target=None,
                 logit_mask=None, segment_ids=None, prompt_ids=None)

Source from the content-addressed store, hash-verified

264
265
266def build_sample(ids, types=None, paddings=None, positions=None, masks=None, label=None, unique_id=None, target=None,
267 logit_mask=None, segment_ids=None, prompt_ids=None):
268 """Convert to numpy and return a sample consumed by the batch producer."""
269
270 ids_np = np.array(ids, dtype=np.int64)
271 sample = {'text': ids_np, 'label': int(label)}
272 if types is not None:
273 types_np = np.array(types, dtype=np.int64)
274 sample['types'] = types_np
275 if paddings is not None:
276 paddings_np = np.array(paddings, dtype=np.int64)
277 sample['padding_mask'] = paddings_np
278 if positions is not None:
279 positions_np = np.array(positions, dtype=np.int64)
280 sample['position'] = positions_np
281 if masks is not None:
282 masks_np = np.array(masks, dtype=np.int64)
283 sample['mask'] = masks_np
284 if target is not None:
285 target_np = np.array(target, dtype=np.int64)
286 sample['target'] = target_np
287 if logit_mask is not None:
288 logit_mask_np = np.array(logit_mask, dtype=np.int64)
289 sample['logit_mask'] = logit_mask_np
290 if segment_ids is not None:
291 segment_ids = np.array(segment_ids, dtype=np.int64)
292 sample['segment_id'] = segment_ids
293 if prompt_ids is not None:
294 prompt_ids = np.array(prompt_ids, dtype=np.int64)
295 sample['prompt_pos'] = prompt_ids
296 if unique_id is not None:
297 sample['uid'] = unique_id
298 return sample
299
300
301def build_decoder_sample(sample, dec_ids, dec_position, dec_masks, dec_target, dec_logit_mask):

Callers 6

encodeMethod · 0.90
encodeMethod · 0.90
__getitem__Method · 0.90
encodeMethod · 0.90
encodeMethod · 0.90
encodeMethod · 0.90

Calls

no outgoing calls

Tested by

no test coverage detected