chenchenghao/sglang
1
0
Fork 0
Code Issues Pull Requests Actions 12 Packages Projects Releases Wiki Activity

[Auto Sync] Update activation.py, logits_processor.py, rota... (20251107) (#12853)

Browse Source 
Co-authored-by: github-actions[bot] <github-actions[bot]@users.noreply.github.com>
Co-authored-by: Stefan He <hebiaobuaa@gmail.com>
This commit is contained in:
Lianmin Zheng
2025-11-07 22:07:51 -08:00
committed by GitHub
co-authored by github-actions[bot] <github-actions[bot]@users.noreply.github.com> Stefan He
parent e039ff382c
commit 0296f1cdad
7 changed files with 21 additions and 15 deletions
Download Patch File Download Diff File Expand all files Collapse all files
python/sglang/srt/models/qwen3.py
+4 -4
View File
@@ -94,7 +94,7 @@ class Qwen3Attention(nn.Module):
weight_dtype=torch.float32,
cast_x_before_out_mul=True,
)
if get_global_server_args().rl_on_policy_target == "fsdp"
if get_global_server_args().rl_on_policy_target is not None
else {}
)
self.q_norm = RMSNorm(self.head_dim, eps=rms_norm_eps, **norm_kwargs)
@@ -167,7 +167,7 @@ class Qwen3Attention(nn.Module):
hidden_states: torch.Tensor,
forward_batch: ForwardBatch,
) -> torch.Tensor:
if get_global_server_args().rl_on_policy_target == "fsdp":
if get_global_server_args().rl_on_policy_target is not None:
hidden_states = hidden_states.bfloat16()
qkv, _ = self.qkv_proj(hidden_states)
@@ -175,7 +175,7 @@ class Qwen3Attention(nn.Module):
q, k = self._apply_qk_norm(q, k)
q, k = self.rotary_emb(positions, q, k)
if get_global_server_args().rl_on_policy_target == "fsdp":
if get_global_server_args().rl_on_policy_target is not None:
q = q.to(torch.bfloat16)
k = k.to(torch.bfloat16)
@@ -229,7 +229,7 @@ class Qwen3DecoderLayer(nn.Module):
override_orig_dtype=torch.float32,
fp32_residual=True,
)
if get_global_server_args().rl_on_policy_target == "fsdp"
if get_global_server_args().rl_on_policy_target is not None
else {}
)
self.input_layernorm = RMSNorm(