[Feature][Kernel]FusedMoE LoRA (#21229)
Signed-off-by: wuchen <cntryroa@gmail.com> Signed-off-by: banjuede <lmklhc@163.com> Signed-off-by: Chen Wu <cntryroa@gmail.com> Signed-off-by: Danielle Robinson <dmmaddix@amazon.com> Signed-off-by: Jee Jee Li <pandaleefree@gmail.com> Signed-off-by: bk-201 <joy25810@foxmail.com> Co-authored-by: wuchen <wuchen@zetyun.com> Co-authored-by: Nathan Van Gheem <vangheem@gmail.com> Co-authored-by: banjuede <lmklhc@163.com> Co-authored-by: Danielle Robinson <dmmaddix@amazon.com> Co-authored-by: Jee Jee Li <pandaleefree@gmail.com> Co-authored-by: bk-201 <joy25810@foxmail.com>
This commit is contained in:
co-authored by
wuchen
Nathan Van Gheem
banjuede
Danielle Robinson
Jee Jee Li
bk-201
parent
3ada34f9cb
commit
5f6cbf60d6
@@ -1313,6 +1313,17 @@ class DeepseekV2ForCausalLM(nn.Module, SupportsPP, MixtureOfExperts, SupportsLoR
|
||||
logits = self.logits_processor(self.lm_head, hidden_states)
|
||||
return logits
|
||||
|
||||
def get_expert_mapping(self) -> list[tuple[str, str, int, str]]:
|
||||
# Params for weights, fp8 weight scales, fp8 activation scales
|
||||
# (param_name, weight_name, expert_id, shard_id)
|
||||
return SharedFusedMoE.make_expert_params_mapping(
|
||||
ckpt_gate_proj_name="gate_proj",
|
||||
ckpt_down_proj_name="down_proj",
|
||||
ckpt_up_proj_name="up_proj",
|
||||
num_experts=self.config.n_routed_experts,
|
||||
num_redundant_experts=0,
|
||||
)
|
||||
|
||||
def load_weights(self, weights: Iterable[tuple[str, torch.Tensor]]) -> set[str]:
|
||||
stacked_params_mapping = [
|
||||
# (param_name, shard_name, shard_id)
|
||||
|
||||
Reference in New Issue
Block a user