[Fix] Put VoxCPM2 hf_overrides under engine_extras

Signed-off-by: Sy03 <1370724210@qq.com>
2026-07-04 16:58:48 +08:00
1 changed files with 15 additions and 14 deletions
@@ -28,20 +28,21 @@ stages:
    max_num_batched_tokens: 4096
    max_model_len: 4096
    devices: "0"
-    hf_overrides:
+    engine_extras:
-      voxcpm2_runtime_config:
+      hf_overrides:
-        audio_emit_every: 1
+        voxcpm2_runtime_config:
-        vae_decode_every: 3
+          audio_emit_every: 1
-        enable_batched_cfm: true
+          vae_decode_every: 3
-        enable_batched_vae_decode: true
+          enable_batched_cfm: true
-        coalesce_audio_d2h: true
+          enable_batched_vae_decode: true
-        enable_loc_dit_fused_qkv: true
+          coalesce_audio_d2h: true
-        enable_loc_dit_fused_mlp: true
+          enable_loc_dit_fused_qkv: true
-        enable_loc_dit_zero_dt_cache: true
+          enable_loc_dit_fused_mlp: true
-        enable_loc_dit_skip_qkv_contig: true
+          enable_loc_dit_zero_dt_cache: true
-        enable_loc_dit_reduce_overhead_no_cg: true
+          enable_loc_dit_skip_qkv_contig: true
-        enable_unified_decode_graph: true
+          enable_loc_dit_reduce_overhead_no_cg: true
-        unified_decode_graph_max_batch_size: 8
+          enable_unified_decode_graph: true
          unified_decode_graph_max_batch_size: 8
    default_sampling_params:
      temperature: 0.0
      top_p: 1.0