Update clap_wrapper.py

clap_wrapper.py

@@ -1,35 +1,14 @@
 import sys
-import os
-
 import torch
 from transformers import ClapModel, ClapProcessor
-from huggingface_hub import hf_hub_download
-from config import config
-
-# กำหนดชื่อและ path ของโมเดล
-HF_REPO_ID = "laion/clap-htsat-fused"
-LOCAL_PATH = "./emotional/clap-htsat-fused"
 
-# ตรวจสอบว่ามีไฟล์โมเดลใน LOCAL_PATH แล้วหรือยัง ถ้าไม่มีก็ดาวน์โหลด
-def ensure_model_downloaded():
-    os.makedirs(LOCAL_PATH, exist_ok=True)
-    required_files = ["pytorch_model.bin", "config.json", "preprocessor_config.json"]
-    for file in required_files:
-        local_file_path = os.path.join(LOCAL_PATH, file)
-        if not os.path.isfile(local_file_path):
-            print(f"Downloading {file} from {HF_REPO_ID}...")
-            hf_hub_download(
-                repo_id=HF_REPO_ID,
-                filename=file,
-                cache_dir=LOCAL_PATH,
-                force_download=False
-            )
-
-ensure_model_downloaded()
+from config import config
 
-# โหลด processor
+# ใช้โมเดลจาก Hugging Face Hub
+REPO_NAME = "laion/clap-htsat-fused"
 models = dict()
-processor = ClapProcessor.from_pretrained("laion/clap-htsat-fused")
+processor = ClapProcessor.from_pretrained(REPO_NAME)
+
 
 def get_clap_audio_feature(audio_data, device=config.bert_gen_config.device):
     if (
@@ -43,12 +22,10 @@ def get_clap_audio_feature(audio_data, device=config.bert_gen_config.device):
     if device not in models.keys():
         if config.webui_config.fp16_run:
             models[device] = ClapModel.from_pretrained(
-                LOCAL_PATH, torch_dtype=torch.float16, local_files_only=True
+                REPO_NAME, torch_dtype=torch.float16
             ).to(device)
         else:
-            models[device] = ClapModel.from_pretrained(
-                LOCAL_PATH, local_files_only=True
-            ).to(device)
+            models[device] = ClapModel.from_pretrained(REPO_NAME).to(device)
     with torch.no_grad():
         inputs = processor(
             audios=audio_data, return_tensors="pt", sampling_rate=48000
@@ -69,12 +46,10 @@ def get_clap_text_feature(text, device=config.bert_gen_config.device):
     if device not in models.keys():
         if config.webui_config.fp16_run:
             models[device] = ClapModel.from_pretrained(
-                LOCAL_PATH, torch_dtype=torch.float16, local_files_only=True
+                REPO_NAME, torch_dtype=torch.float16
             ).to(device)
         else:
-            models[device] = ClapModel.from_pretrained(
-                LOCAL_PATH, local_files_only=True
-            ).to(device)
+            models[device] = ClapModel.from_pretrained(REPO_NAME).to(device)
     with torch.no_grad():
         inputs = processor(text=text, return_tensors="pt").to(device)
         emb = models[device].get_text_features(**inputs).float()