[Bugfix] Propagate quant_config to LFM2 ShortConv projections (#48917)
Signed-off-by: Alex Yuan <alex.yuan@liquid.ai> Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
This commit is contained in:
co-authored by
Claude Opus 4.8
parent
616c9bd0f4
commit
8def3cdde2
@@ -257,6 +257,7 @@ class Lfm2ShortConvDecoderLayer(nn.Module):
|
||||
layer_idx=layer_idx,
|
||||
model_config=model_config,
|
||||
cache_config=cache_config,
|
||||
quant_config=quant_config,
|
||||
prefix=f"{prefix}.conv",
|
||||
)
|
||||
|
||||
|
||||
@@ -351,6 +351,7 @@ class Lfm2MoeShortConvDecoderLayer(nn.Module):
|
||||
layer_idx=layer_idx,
|
||||
model_config=model_config,
|
||||
cache_config=cache_config,
|
||||
quant_config=quant_config,
|
||||
prefix=f"{prefix}.conv",
|
||||
)
|
||||
|
||||
|
||||
Reference in New Issue
Block a user