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

Function load_weights

utils.py:414–431  ·  view source on GitHub ↗

Loads weights from src to dst via in place copy. src is a huggingface gpt2model, while dst is one of our models. dst2src=True loads parameters from our models into huggingface's. ^dst2src is still untested

(src, dst, dst2src=False)

Source from the content-addressed store, hash-verified

412
413
414def load_weights(src, dst, dst2src=False):
415 """
416 Loads weights from src to dst via in place copy.
417 src is a huggingface gpt2model, while dst is one of our models.
418 dst2src=True loads parameters from our models into huggingface's.
419 ^dst2src is still untested
420 """
421 conv_layer = 'Conv1D' in str(type(src))
422 for n, p in src.named_parameters():
423 if dst2src:
424 data = dst._parameters[n].data
425 load = p.data
426 else:
427 data = p.data
428 load = dst._parameters[n].data
429 if conv_layer and 'weight' in n:
430 data = data.t().contiguous()
431 load.copy_(data)
432
433
434# dst._parameters[n].data.copy_(data)

Callers 4

load_mlpFunction · 0.85
load_attentionFunction · 0.85
load_transformer_layerFunction · 0.85
move_weightsFunction · 0.85

Calls 1

named_parametersMethod · 0.45

Tested by

no test coverage detected