Skip to content

[NO-TICKET] Add fused_rms_norm config usage in Mistral moreh #43

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
codernine-moreh merged 2 commits into from
Oct 31, 2025
Merged

[NO-TICKET] Add fused_rms_norm config usage in Mistral moreh #43

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
f71559b
Add fused_rms_norm usage
codernine-moreh Oct 30, 2025
6273096
Fix term
codernine-moreh Oct 31, 2025
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
29 changes: 26 additions & 3 deletions src/transformers/models/mistral/modeling_mistral_moreh.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -56,11 +56,21 @@

_flash_supports_window_size = "window_size" in list(inspect.signature(flash_attn_func).parameters)

try:
# Moreh extension
moreh_ops = torch.ops.moreh
MorehRMSNorm = moreh_ops.T5LayerNorm
except AttributeError:
MorehRMSNorm = None

logger = logging.get_logger(__name__)

_CONFIG_FOR_DOC = "MistralConfig"

if MorehRMSNorm is not None:
logger.warning(
"You can't use Masked Structured Growth Training..! You should avoid using RMSNorm in any way. "
)

# Copied from transformers.models.llama.modeling_llama._get_unpad_data
def _get_unpad_data(attention_mask):
Expand All @@ -74,6 +84,12 @@ def _get_unpad_data(attention_mask):
max_seqlen_in_batch,
)

def get_moreh_fused_rms_norm(config):
moreh_config = getattr(config, "moreh_config", None)
if moreh_config is not None and "fused_rms_norm" in moreh_config:
return moreh_config["fused_rms_norm"]
return False


# Copied from transformers.models.llama.modeling_llama.LlamaRMSNorm with Llama->Mistral
class MistralRMSNorm(nn.Module):
Expand Down Expand Up @@ -710,8 +726,12 @@ def __init__(self, config: MistralMorehConfig, layer_idx: int):
self.self_attn = MISTRAL_ATTENTION_CLASSES[config._attn_implementation](config=config, layer_idx=layer_idx)

self.mlp = MistralMLP(config)
self.input_layernorm = MistralRMSNorm(config.hidden_size, eps=config.rms_norm_eps)
self.post_attention_layernorm = MistralRMSNorm(config.hidden_size, eps=config.rms_norm_eps)
if get_moreh_fused_rms_norm(config):
self.input_layernorm = MorehRMSNorm(config.hidden_size, eps=config.rms_norm_eps)
self.post_attention_layernorm = MorehRMSNorm(config.hidden_size, eps=config.rms_norm_eps)
else:
self.input_layernorm = MistralRMSNorm(config.hidden_size, eps=config.rms_norm_eps)
self.post_attention_layernorm = MistralRMSNorm(config.hidden_size, eps=config.rms_norm_eps)

def forward(
self,
Expand Down Expand Up @@ -912,7 +932,10 @@ def __init__(self, config: MistralMorehConfig):
[MistralDecoderLayer(config, layer_idx) for layer_idx in range(config.num_hidden_layers)]
)
self._attn_implementation = config._attn_implementation
self.norm = MistralRMSNorm(config.hidden_size, eps=config.rms_norm_eps)
if get_moreh_fused_rms_norm(config):
self.norm = MorehRMSNorm(config.hidden_size, eps=config.rms_norm_eps)
else:
self.norm = MistralRMSNorm(config.hidden_size, eps=config.rms_norm_eps)

self.gradient_checkpointing = False
# Initialize weights and apply final processing
Expand Down