diff --git a/api_server.py b/api_server.py
new file mode 100644
index 0000000..51a4e0a
--- /dev/null
+++ b/api_server.py
@@ -0,0 +1,134 @@
+#!/usr/bin/env python3
+import os
+import sys
+
+# 关键：强制离线模式，避免连接 HuggingFace
+os.environ["HF_HUB_OFFLINE"] = "1"
+os.environ["TRANSFORMERS_OFFLINE"] = "1"
+os.environ["HF_DATASETS_OFFLINE"] = "1"
+
+import traceback
+from pathlib import Path
+
+def log_error(msg):
+    print(f"[ERROR] {msg}", file=sys.stderr)
+
+def log_info(msg):
+    print(f"[INFO] {msg}")
+
+try:
+    log_info("Starting API server (OFFLINE MODE)...")
+    log_info(f"Python version: {sys.version}")
+    
+    from fastapi import FastAPI, UploadFile, File, Form
+    from fastapi.responses import FileResponse
+    import uvicorn
+    import tempfile
+    
+    log_info("Dependencies imported successfully")
+    
+    sys.path.insert(0, os.path.dirname(os.path.abspath(__file__)))
+    from indextts.infer_v2 import IndexTTS2
+    from transformers import SeamlessM4TFeatureExtractor
+    
+    log_info("IndexTTS2 imported successfully")
+    
+    # 检查模型路径
+    cfg_path = "checkpoints/config.yaml"
+    model_dir = "checkpoints"
+    
+    if not os.path.exists(cfg_path):
+        log_error(f"Config file not found: {cfg_path}")
+        sys.exit(1)
+    
+    log_info("Initializing model...")
+    
+    # DGX Spark 优化：禁用 CUDA 内核编译（因为 sm_121 不支持 compute_70）
+    # 强制使用纯 PyTorch 模式
+    os.environ["DISABLE_BIGVGAN_CUDA_KERNEL"] = "1"
+    
+    local_w2v_path = "./models/w2v-bert-2.0"
+    
+    if os.path.exists(local_w2v_path):
+        log_info(f"Loading w2v-bert-2.0 from local path: {local_w2v_path}")
+        feature_extractor = SeamlessM4TFeatureExtractor.from_pretrained(local_w2v_path)
+        log_info("Feature extractor loaded locally")
+    else:
+        log_info("Local w2v-bert-2.0 not found, will try to load from cache...")
+    
+    # 初始化 IndexTTS2
+    try:
+        tts = IndexTTS2(
+            cfg_path=cfg_path,
+            model_dir=model_dir,
+            use_fp16=True,  # DGX Spark 支持 FP16
+            use_cuda_kernel=False  # 关键：禁用自定义 CUDA 内核，避免 sm_121 编译错误
+        )
+        log_info("Model loaded successfully!")
+    except Exception as e:
+        log_error(f"Model initialization failed: {e}")
+        traceback.print_exc()
+        sys.exit(1)
+    
+    app = FastAPI(title="IndexTTS2 API (DGX Spark)")
+    
+    @app.post("/tts")
+    async def text_to_speech(
+        text: str = Form(...),
+        spk_audio: UploadFile = File(...),
+        emo_audio: UploadFile = None,
+        emo_alpha: float = Form(0.8)
+    ):
+        log_info(f"Processing: {text[:50]}...")
+        
+        with tempfile.NamedTemporaryFile(suffix=".wav", delete=False) as spk_tmp:
+            content = await spk_audio.read()
+            spk_tmp.write(content)
+            spk_path = spk_tmp.name
+    
+        emo_path = None
+        if emo_audio:
+            with tempfile.NamedTemporaryFile(suffix=".wav", delete=False) as emo_tmp:
+                content = await emo_audio.read()
+                emo_tmp.write(content)
+                emo_path = emo_tmp.name
+    
+        output_path = tempfile.mktemp(suffix=".wav")
+        
+        try:
+            tts.infer(
+                spk_audio_prompt=spk_path,
+                text=text,
+                output_path=output_path,
+                emo_audio_prompt=emo_path,
+                emo_alpha=emo_alpha
+            )
+            
+            return FileResponse(
+                output_path, 
+                media_type="audio/wav",
+                filename="generated.wav"
+            )
+        finally:
+            if os.path.exists(spk_path):
+                os.unlink(spk_path)
+            if emo_path and os.path.exists(emo_path):
+                os.unlink(emo_path)
+    
+    @app.get("/health")
+    def health_check():
+        return {"status": "ok", "model": "IndexTTS2", "device": "cuda"}
+    
+    @app.get("/")
+    def root():
+        return {"message": "IndexTTS2 API for DGX Spark", "docs": "/docs"}
+    
+    log_info("Starting server on port 11996...")
+    
+    if __name__ == "__main__":
+        uvicorn.run(app, host="0.0.0.0", port=11996)
+        
+except Exception as e:
+    log_error(f"Fatal error: {e}")
+    traceback.print_exc()
+    sys.exit(1)
\ No newline at end of file
diff --git a/indextts/infer_v2.py b/indextts/infer_v2.py
index c5b85fb..c5415d6 100644
--- a/indextts/infer_v2.py
+++ b/indextts/infer_v2.py
@@ -113,7 +113,7 @@ class IndexTTS2:
                 print(f"{e!r}")
                 self.use_cuda_kernel = False
 
-        self.extract_features = SeamlessM4TFeatureExtractor.from_pretrained("facebook/w2v-bert-2.0")
+        self.extract_features = SeamlessM4TFeatureExtractor.from_pretrained("./models/w2v-bert-2.0")
         self.semantic_model, self.semantic_mean, self.semantic_std = build_semantic_model(
             os.path.join(self.model_dir, self.cfg.w2v_stat))
         self.semantic_model = self.semantic_model.to(self.device)
@@ -122,7 +122,8 @@ class IndexTTS2:
         self.semantic_std = self.semantic_std.to(self.device)
 
         semantic_codec = build_semantic_codec(self.cfg.semantic_codec)
-        semantic_code_ckpt = hf_hub_download("amphion/MaskGCT", filename="semantic_codec/model.safetensors")
+        # semantic_code_ckpt = hf_hub_download("./models/MaskGCT", filename="./models/MaskGCT/semantic_codec/model.safetensors")
+        semantic_code_ckpt = "./models/MaskGCT/semantic_codec/model.safetensors"
         safetensors.torch.load_model(semantic_codec, semantic_code_ckpt)
         self.semantic_codec = semantic_codec.to(self.device)
         self.semantic_codec.eval()
@@ -151,9 +152,8 @@ class IndexTTS2:
         print(">> s2mel weights restored from:", s2mel_path)
 
         # load campplus_model
-        campplus_ckpt_path = hf_hub_download(
-            "funasr/campplus", filename="campplus_cn_common.bin"
-        )
+        # campplus_ckpt_path = hf_hub_download("funasr/campplus", filename="campplus_cn_common.bin")
+        campplus_ckpt_path = "./models/campplus/campplus_cn_common.bin"
         campplus_model = CAMPPlus(feat_dim=80, embedding_size=192)
         campplus_model.load_state_dict(torch.load(campplus_ckpt_path, map_location="cpu"))
         self.campplus_model = campplus_model.to(self.device)
diff --git a/indextts/s2mel/modules/bigvgan/alias_free_activation/cuda/build/.ninja_deps b/indextts/s2mel/modules/bigvgan/alias_free_activation/cuda/build/.ninja_deps
index aae7506..e024e2d 100644
Binary files a/indextts/s2mel/modules/bigvgan/alias_free_activation/cuda/build/.ninja_deps and b/indextts/s2mel/modules/bigvgan/alias_free_activation/cuda/build/.ninja_deps differ
diff --git a/indextts/s2mel/modules/bigvgan/alias_free_activation/cuda/build/.ninja_log b/indextts/s2mel/modules/bigvgan/alias_free_activation/cuda/build/.ninja_log
index 773d44e..b1be289 100644
--- a/indextts/s2mel/modules/bigvgan/alias_free_activation/cuda/build/.ninja_log
+++ b/indextts/s2mel/modules/bigvgan/alias_free_activation/cuda/build/.ninja_log
@@ -1,2 +1,3 @@
 # ninja log v7
 0	8799	1771051743097851885	anti_alias_activation.o	6d4acaa78fc0c336
+0	9806	1771307045391159629	anti_alias_activation.o	6d4acaa78fc0c336
diff --git a/indextts/utils/maskgct_utils.py b/indextts/utils/maskgct_utils.py
index 40b9cb0..99c0e07 100644
--- a/indextts/utils/maskgct_utils.py
+++ b/indextts/utils/maskgct_utils.py
@@ -85,7 +85,7 @@ class JsonHParams:
 
 
 def build_semantic_model(path_='./models/tts/maskgct/ckpt/wav2vec2bert_stats.pt'):
-    semantic_model = Wav2Vec2BertModel.from_pretrained("facebook/w2v-bert-2.0")
+    semantic_model = Wav2Vec2BertModel.from_pretrained("./models/w2v-bert-2.0")
     semantic_model.eval()
     stat_mean_var = torch.load(path_)
     semantic_mean = stat_mean_var["mean"]