remove: T4 auto-optimizations detection — use standard Applio/VRVC approach

Neither Applio nor Vietnamese-RVC have T4-specific detection or
optimizations. They just use standard CUDA with cache_data_in_gpu
as a user option. Our T4 detection added unnecessary special-casing
that didn't improve training quality.

Removed:
- _is_t4 detection block in train.py
- T4 log message in training startup
- T4-specific GPU class reporting in cli.py (show_info + version)
- T4 mention in README Colab table

Kept: ZLUDA detection (this is genuinely needed for AMD GPU support)

Author: BF667

README.md

@@ -102,7 +102,7 @@ rvc-cli --help
 
 | Notebook | Description |
 |----------|-------------|
-| [![Open in Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/ArkanDash/Advanced-RVC-Inference/blob/master/Advanced-RVC.ipynb) | Full Web UI with T4 auto-optimizations |
+| [![Open in Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/ArkanDash/Advanced-RVC-Inference/blob/master/Advanced-RVC.ipynb) | Full Web UI |
 | [![Open in Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/ArkanDash/Advanced-RVC-Inference/blob/master/colab-noui.ipynb) | CLI only — lightweight headless mode |
 
 ---

arvc/api/cli.py

@@ -103,16 +103,13 @@ def show_version():
             version_info.append(f"CUDA: {torch.version.cuda}")
             gpu_mem = torch.cuda.get_device_properties(0).total_memory // (1024**3)
             version_info.append(f"GPU Memory: {gpu_mem} GB")
-            # ZLUDA / T4 detection
+            # ZLUDA detection
             try:
                 from arvc.engine.models.backends import zluda
                 if zluda.is_available():
                     version_info.append(f"ZLUDA: Detected (AMD GPU via CUDA compatibility)")
             except ImportError:
                 pass
-            gpu_name = torch.cuda.get_device_name(0).lower()
-            if "t4" in gpu_name or "tesla t4" in gpu_name:
-                version_info.append(f"GPU: Tesla T4 (T4 optimizations active)")
     except ImportError:
         version_info.append("PyTorch: Not installed")
 
@@ -163,12 +160,7 @@ def show_info():
                     info.append(f"  Backend: HIP/ROCm (via ZLUDA)")
             except ImportError:
                 pass
-            # T4 detection
-            gpu_name_lower = gpu_name.lower()
-            if "t4" in gpu_name_lower or "tesla t4" in gpu_name_lower:
-                info.append(f"  GPU Class: Tesla T4 (T4-optimized training defaults)")
-            else:
-                info.append(f"  GPU Class: Standard ({gpu_mem}GB)")
+            info.append(f"  GPU Class: Standard ({gpu_mem}GB)")
         else:
             info.append("  CUDA Available: False")
     except ImportError:

arvc/engine/training/runner/train.py

@@ -45,14 +45,6 @@
 # ZLUDA detection: True when running on AMD GPU via CUDA compatibility layer
 _is_zluda = zluda.is_available()
 
-# T4 detection for Colab-friendly defaults
-_is_t4 = False
-try:
-    if torch.cuda.is_available():
-        _gpu_name = torch.cuda.get_device_name(0).lower()
-        _is_t4 = "t4" in _gpu_name or "tesla t4" in _gpu_name
-except Exception:
-    pass
 from arvc.utils.variables import logger, translations as _raw_translations
 
 # ── BULLETPROOF SAFETY NET ──
@@ -543,8 +535,6 @@ def run(
     if rank == 0:
         if _is_zluda:
             logger.info(f"Training on ZLUDA (AMD GPU): {zluda.device_name(0)}")
-        elif _is_t4:
-            logger.info(f"Training on T4 GPU — using T4-optimized defaults (FP16, grad accumulation)")
 
     writer_eval = SummaryWriter(
         log_dir=eval_dir