Fix issue where errors on qwen image omni were not being fully surfaced.
This commit is contained in:
@@ -760,6 +760,13 @@ class Qwen3OmniCaptioner(BaseCaptioner):
|
|||||||
else:
|
else:
|
||||||
inputs = self._process_items(items)
|
inputs = self._process_items(items)
|
||||||
inputs = inputs.to(self.device_torch).to(self.torch_dtype)
|
inputs = inputs.to(self.device_torch).to(self.torch_dtype)
|
||||||
|
# a generate that dies between static-cache creation and its first
|
||||||
|
# forward leaves model._cache with uninitialized layers; transformers
|
||||||
|
# then raises AttributeError reading cache.max_batch_size on every
|
||||||
|
# later call, masking the original error — drop the stale cache
|
||||||
|
stale_cache = getattr(self.model, "_cache", None)
|
||||||
|
if stale_cache is not None and not stale_cache.is_initialized:
|
||||||
|
del self.model._cache
|
||||||
# under static cache, generate hands the forward a prepared 4D mask;
|
# under static cache, generate hands the forward a prepared 4D mask;
|
||||||
# the true 2D padding mask is needed for the prefill rope index
|
# the true 2D padding mask is needed for the prefill rope index
|
||||||
self.model._pad_mask_2d = inputs.get("attention_mask", None)
|
self.model._pad_mask_2d = inputs.get("attention_mask", None)
|
||||||
|
|||||||
Reference in New Issue
Block a user