Co-authored-by: JD <jaedon.guo@gmail.com>
This commit is contained in:
@@ -38,7 +38,7 @@ class WeightChecker:
|
|||||||
|
|
||||||
def _reset_tensors(self):
|
def _reset_tensors(self):
|
||||||
for name, param in self._model_state():
|
for name, param in self._model_state():
|
||||||
if "cos_sin_cache" in name or "freqs_cis" in name:
|
if "cos_sin_cache" in name or "freqs_cis" in name or "_weight_fp32" in name:
|
||||||
continue
|
continue
|
||||||
param.copy_(_random_like(param))
|
param.copy_(_random_like(param))
|
||||||
|
|
||||||
@@ -131,6 +131,7 @@ def _postprocess_tensors(
|
|||||||
non_persistent_buffer_patterns = [
|
non_persistent_buffer_patterns = [
|
||||||
"cos_sin_cache", # RoPE cache
|
"cos_sin_cache", # RoPE cache
|
||||||
"inv_freq", # RoPE inverse frequency (if it exists as buffer)
|
"inv_freq", # RoPE inverse frequency (if it exists as buffer)
|
||||||
|
"_weight_fp32", # FP32 cache of gate weight (e.g. Glm4MoeGate)
|
||||||
]
|
]
|
||||||
|
|
||||||
for name in raw:
|
for name in raw:
|
||||||
|
|||||||
Reference in New Issue
Block a user