Length masking for batch inputs (#1173)

* length masking

* add mask to mlx_lm model interface

* remove lengths

* fix test:

* comment + fix
This commit is contained in:
Alex Barron
2024-12-18 19:43:52 -08:00
committed by GitHub
parent db109184b7
commit d4ef909d4a
34 changed files with 191 additions and 72 deletions

View File

@@ -176,11 +176,13 @@ class NemotronModel(nn.Module):
def __call__(
self,
inputs: mx.array,
mask: mx.array = None,
cache=None,
):
h = self.embed_tokens(inputs)
mask = create_attention_mask(h, cache)
if mask is None:
mask = create_attention_mask(h, cache)
if cache is None:
cache = [None] * len(self.layers)
@@ -203,9 +205,10 @@ class Model(nn.Module):
def __call__(
self,
inputs: mx.array,
mask: mx.array = None,
cache=None,
):
out = self.model(inputs, cache)
out = self.model(inputs, mask, cache)
if self.args.tie_word_embeddings:
out = self.model.embed_tokens.as_linear(out)
else: