Skip to content

Add a way to disable the final norm in the llama based TE models. #10794

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
comfyanonymous merged 1 commit into from
Nov 19, 2025
Merged

Add a way to disable the final norm in the llama based TE models. #10794

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
18 changes: 15 additions & 3 deletions comfy/text_encoders/llama.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -32,6 +32,7 @@ class Llama2Config:
q_norm = None
k_norm = None
rope_scale = None
final_norm: bool = True

@dataclass
class Qwen25_3BConfig:
Expand All @@ -53,6 +54,7 @@ class Qwen25_3BConfig:
q_norm = None
k_norm = None
rope_scale = None
final_norm: bool = True

@dataclass
class Qwen25_7BVLI_Config:
Expand All @@ -74,6 +76,7 @@ class Qwen25_7BVLI_Config:
q_norm = None
k_norm = None
rope_scale = None
final_norm: bool = True

@dataclass
class Gemma2_2B_Config:
Expand All @@ -96,6 +99,7 @@ class Gemma2_2B_Config:
k_norm = None
sliding_attention = None
rope_scale = None
final_norm: bool = True

@dataclass
class Gemma3_4B_Config:
Expand All @@ -118,6 +122,7 @@ class Gemma3_4B_Config:
k_norm = "gemma3"
sliding_attention = [False, False, False, False, False, 1024]
rope_scale = [1.0, 8.0]
final_norm: bool = True

class RMSNorm(nn.Module):
def __init__(self, dim: int, eps: float = 1e-5, add=False, device=None, dtype=None):
Expand Down Expand Up @@ -366,7 +371,12 @@ def __init__(self, config, device=None, dtype=None, ops=None):
transformer(config, index=i, device=device, dtype=dtype, ops=ops)
for i in range(config.num_hidden_layers)
])
self.norm = RMSNorm(config.hidden_size, eps=config.rms_norm_eps, add=config.rms_norm_add, device=device, dtype=dtype)

if config.final_norm:
self.norm = RMSNorm(config.hidden_size, eps=config.rms_norm_eps, add=config.rms_norm_add, device=device, dtype=dtype)
else:
self.norm = None

# self.lm_head = ops.Linear(config.hidden_size, config.vocab_size, bias=False, device=device, dtype=dtype)

def forward(self, x, attention_mask=None, embeds=None, num_tokens=None, intermediate_output=None, final_layer_norm_intermediate=True, dtype=None, position_ids=None, embeds_info=[]):
Expand Down Expand Up @@ -421,14 +431,16 @@ def forward(self, x, attention_mask=None, embeds=None, num_tokens=None, intermed
if i == intermediate_output:
intermediate = x.clone()

x = self.norm(x)
if self.norm is not None:
x = self.norm(x)

if all_intermediate is not None:
all_intermediate.append(x.unsqueeze(1).clone())

if all_intermediate is not None:
intermediate = torch.cat(all_intermediate, dim=1)

if intermediate is not None and final_layer_norm_intermediate:
if intermediate is not None and final_layer_norm_intermediate and self.norm is not None:
intermediate = self.norm(intermediate)

return x, intermediate
Expand Down