[VLM] feat: true on policy for vlm + fsdp (#14636)
Co-authored-by: Xinyuan Tong <115166877+JustinTong0323@users.noreply.github.com>
This commit is contained in:
co-authored by
Xinyuan Tong
parent
00562ee14a
commit
7254986342
@@ -565,11 +565,25 @@ class VisionAttention(nn.Module):
|
||||
self.dummy_dim = (num_dummy_heads + num_heads) * self.head_size
|
||||
|
||||
if self.qk_normalization:
|
||||
norm_kwargs = (
|
||||
dict(
|
||||
weight_dtype=torch.float32,
|
||||
cast_x_before_out_mul=True,
|
||||
)
|
||||
if get_global_server_args().rl_on_policy_target is not None
|
||||
else {}
|
||||
)
|
||||
self.q_norm = RMSNorm(
|
||||
self.dummy_dim, eps=layer_norm_eps, var_hidden_size=embed_dim
|
||||
self.dummy_dim,
|
||||
eps=layer_norm_eps,
|
||||
var_hidden_size=embed_dim,
|
||||
**norm_kwargs,
|
||||
)
|
||||
self.k_norm = RMSNorm(
|
||||
self.dummy_dim, eps=layer_norm_eps, var_hidden_size=embed_dim
|
||||
self.dummy_dim,
|
||||
eps=layer_norm_eps,
|
||||
var_hidden_size=embed_dim,
|
||||
**norm_kwargs,
|
||||
)
|
||||
|
||||
# Select attention backend via a unified method
|
||||
@@ -720,6 +734,15 @@ class VisionAttention(nn.Module):
|
||||
if x.dim() == 2:
|
||||
x = x.unsqueeze(0)
|
||||
assert x.dim() == 3, x.shape
|
||||
if (
|
||||
get_global_server_args().rl_on_policy_target is not None
|
||||
and position_embeddings is not None
|
||||
):
|
||||
assert isinstance(position_embeddings, tuple), (
|
||||
"expected position_embeddings to be a tuple of two tensors,\n"
|
||||
f"but got {type(position_embeddings)}, change if needed"
|
||||
)
|
||||
position_embeddings = tuple(p.to(x.dtype) for p in position_embeddings)
|
||||
x_shape = x.shape
|
||||
bsz, s, _ = x_shape
|
||||
head = self.num_attention_heads_per_partition
|
||||
|
||||
Reference in New Issue
Block a user