apps/FastDeploy
1
0
Fork 0
mirror of https://github.com/PaddlePaddle/FastDeploy.git synced 2025-10-05 08:37:06 +08:00
Code Issues Actions 2 Packages Projects Releases Wiki Activity

refactor rl get_name_mappings_to_training (#2847)
Some checks failed
Deploy GitHub Pages / deploy (push) Has been cancelled
Details

Browse Source 
* refactor rl get_name_mappings_to_training

* fix tp>1

* change variable name(ffn1->up_gate_proj/ffn2->down_proj)

* change variable name(linear_weight->weight/linear_bias->bias)

* add rl names mapping for vl

* fix ernie 0.3B error

* fix develop code

* fix
This commit is contained in:
Yuanle Liu
2025-07-15 22:31:42 +08:00
committed by GitHub
parent e7bcbbab52
commit 61b3997b85
47 changed files with 1591 additions and 1629 deletions
Download Patch File Download Diff File Expand all files Collapse all files

10
fastdeploy/model_executor/models/qwen3.py
View File

@@ -166,7 +166,7 @@ class Qwen3Model(nn.Layer):
self.num_layers = fd_config.model_config.num_hidden_layers
fd_config.model_config.pretrained_config.prefix_name = "model"
self.embeddings = VocabParallelEmbedding(
self.embed_tokens = VocabParallelEmbedding(
fd_config=fd_config,
num_embeddings=fd_config.model_config.vocab_size,
embedding_dim=fd_config.model_config.hidden_size,
@@ -197,7 +197,7 @@ class Qwen3Model(nn.Layer):
A dictionary containing model parameters, where keys are parameter names
and values are NumPy arrays or PaddlePaddle tensors.
"""
self.embeddings.load_state_dict(state_dict)
self.embed_tokens.load_state_dict(state_dict)
self.norm.load_state_dict(state_dict)
for i in range(self.num_layers):
logger.info(f"Start load layer {i}")
@@ -210,7 +210,7 @@ class Qwen3Model(nn.Layer):
):
"""
"""
hidden_states = self.embeddings(ids_remove_padding=ids_remove_padding)
hidden_states = self.embed_tokens(ids_remove_padding=ids_remove_padding)
residual = None
@@ -266,8 +266,8 @@ class Qwen3ForCausalLM(ModelForCasualLM):
"""
self.model.load_state_dict(state_dict)
if self.tie_word_embeddings:
self.lm_head.out_linear.weight.set_value(
self.model.embeddings.word_embeddings.weight.transpose([1, 0]))
self.lm_head.linear.weight.set_value(
self.model.embed_tokens.embeddings.weight.transpose([1, 0]))
else:
self.lm_head.load_state_dict(state_dict)