Remove KV cache scale boilerplate from model weight loading methods (#43167)
Signed-off-by: Harry Mellor <19981378+hmellor@users.noreply.github.com> Co-authored-by: mergify[bot] <37929162+mergify[bot]@users.noreply.github.com>
This commit is contained in:
co-authored by
mergify[bot] <37929162+mergify[bot]@users.noreply.github.com>
parent
7fe7800fa4
commit
62215e72c6
@@ -395,18 +395,6 @@ class HyperCLOVAXModel(nn.Module):
|
||||
# Models trained using ColossalAI may include these tensors in
|
||||
# the checkpoint. Skip them.
|
||||
continue
|
||||
if self.quant_config is not None and (
|
||||
scale_name := self.quant_config.get_cache_scale(name)
|
||||
):
|
||||
# Loading kv cache quantization scales
|
||||
param = params_dict[scale_name]
|
||||
weight_loader = getattr(param, "weight_loader", default_weight_loader)
|
||||
loaded_weight = (
|
||||
loaded_weight if loaded_weight.dim() == 0 else loaded_weight[0]
|
||||
)
|
||||
weight_loader(param, loaded_weight)
|
||||
loaded_params.add(scale_name)
|
||||
continue
|
||||
if "scale" in name or "zero_point" in name:
|
||||
# Remapping the name of FP8 kv-scale or zero point.
|
||||
remapped_name = maybe_remap_kv_scale_name(name, params_dict)
|
||||
|
||||
Reference in New Issue
Block a user