apps/FastDeploy
1
0
Fork 0
mirror of https://github.com/PaddlePaddle/FastDeploy.git synced 2025-12-24 13:28:13 +08:00
Code Issues Actions 2 Packages Projects Releases Wiki Activity

[TSP] Support qwen3 moe tsp + cudagraph (#4871)
Some checks failed
CE Compile Job / ce_job_pre_check (push) Has been cancelled
Details
CE Compile Job / print_ce_job_pre_check_outputs (push) Has been cancelled
Details
CE Compile Job / FD-Clone-Linux (push) Has been cancelled
Details
CE Compile Job / Show Code Archive Output (push) Has been cancelled
Details
CE Compile Job / BUILD_SM8090 (push) Has been cancelled
Details
CE Compile Job / BUILD_SM8689 (push) Has been cancelled
Details
CE Compile Job / CE_UPLOAD (push) Has been cancelled
Details
Deploy GitHub Pages / deploy (push) Has been cancelled
Details

Browse Source 
* support qwen3_moe tsp mode

* fix

* fix

* update

* update

* update

* fix

* support external_rmsnorm

* update

* fix
This commit is contained in:
Yuanle Liu
2025-11-10 23:37:51 +08:00
committed by GitHub
parent fb2eb403ab
commit 3dc0ffa46d
28 changed files with 173 additions and 273 deletions
Download Patch File Download Diff File Expand all files Collapse all files

21
fastdeploy/model_executor/models/ernie4_5_moe.py
View File

@@ -312,6 +312,7 @@ class Ernie4_5_DecoderLayer(nn.Layer):
hidden_size=fd_config.model_config.hidden_size,
eps=fd_config.model_config.rms_norm_eps,
prefix=f"{prefix}.post_attention_layernorm",
layer_id=layer_id,
)
def load_state_dict(self, state_dict):
@@ -329,18 +330,9 @@ class Ernie4_5_DecoderLayer(nn.Layer):
hidden_states: paddle.Tensor,
residual: paddle.Tensor = None,
):
if residual is None:
residual = hidden_states
hidden_states = self.input_layernorm(
hidden_states,
forward_meta=forward_meta,
)
else:
hidden_states, residual = self.input_layernorm(
hidden_states,
residual,
forward_meta=forward_meta,
)
hidden_states, residual = self.input_layernorm(
hidden_states, residual_input=residual, forward_meta=forward_meta
)
hidden_states = self.self_attn(
hidden_states=hidden_states,
@@ -350,7 +342,6 @@ class Ernie4_5_DecoderLayer(nn.Layer):
hidden_states, residual = self.post_attention_layernorm(
hidden_states,
residual,
forward_meta=forward_meta,
)
hidden_states = self.mlp(hidden_states)
@@ -455,9 +446,7 @@ class Ernie4_5_Model(nn.Layer):
for i in range(self.num_layers):
hidden_states, residual = self.layers[i](forward_meta, hidden_states, residual)
hidden_states = hidden_states + residual
out = self.norm(hidden_states, forward_meta=forward_meta)
out = self.norm(hidden_states, residual, forward_meta=forward_meta)[0]
if current_platform.is_iluvatar() and forward_meta.attn_backend.mixed:
out = forward_meta.attn_backend.reverse_transpose(out)