Function _logps

scripts/train_dpo.py:33–34 · view source on GitHub ↗

(model, ids, mask, requires_grad)

Source from the content-addressed store, hash-verified

31
32
33	def _logps(model, ids, mask, requires_grad):
34	return sequence_logprobs(model, ids, mask, requires_grad=requires_grad)
35
36
37	def _compute_losses(policy, ref, batch, cfg, ctx):

_compute_lossesFunction · 0.85

sequence_logprobsFunction · 0.90

no test coverage detected