From 368afea2f2134132a126027a398d71c091278f45 Mon Sep 17 00:00:00 2001 From: Sy03 <1370724210@qq.com> Date: Sat, 2 May 2026 20:39:32 +0800 Subject: [PATCH] [Bugfix][Fish Speech] Flatten Fast AR RoPE positions (#3299) Signed-off-by: Sy03 <1370724210@qq.com> Co-authored-by: Hongsheng Liu --- .../model_executor/models/fish_speech/fish_speech_fast_ar.py | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/vllm_omni/model_executor/models/fish_speech/fish_speech_fast_ar.py b/vllm_omni/model_executor/models/fish_speech/fish_speech_fast_ar.py index cc38c200a..df3017152 100644 --- a/vllm_omni/model_executor/models/fish_speech/fish_speech_fast_ar.py +++ b/vllm_omni/model_executor/models/fish_speech/fish_speech_fast_ar.py @@ -106,7 +106,7 @@ class _FastARAttention(nn.Module): q = self.q_norm(q.view(-1, self.num_heads, self.head_dim)).view(q.shape) k = self.k_norm(k.view(-1, self.num_kv_heads, self.head_dim)).view(k.shape) - q, k = self.rotary_emb(position_ids, q, k) + q, k = self.rotary_emb(position_ids.reshape(-1), q, k) q = q.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2) k = k.view(bsz, seq_len, self.num_kv_heads, self.head_dim).transpose(1, 2)