Revert "feat: add Orpheus TTS benchmark results (RTX 4090)"

This reverts commit 7065ca304d2b9b13a9cf7fa00ebe21a206abbc70.

.gitattributes

@@ -4,4 +4,3 @@
 *.pth filter=lfs diff=lfs merge=lfs -text
 *.safetensors filter=lfs diff=lfs merge=lfs -text
 *.onnx filter=lfs diff=lfs merge=lfs -text
-*.wav filter=lfs diff=lfs merge=lfs -text

orpheus-tts/2024-12-24/README.md

@@ -1,45 +0,0 @@
-# Orpheus TTS - Teste RTX 4090
-
-**Data:** 2024-12-24
-
-## Configuracao
-
-- **GPU:** NVIDIA RTX 4090 (24GB VRAM)
-- **Modelo:** `canopylabs/orpheus-3b-0.1-ft` (3B parametros)
-- **Precisao:** bfloat16
-- **Backend:** vLLM v0.13.0 com FLASH_ATTN
-- **Plataforma:** Vast.ai
-
-## Resultados
-
-| Metrica | Valor |
-|---------|-------|
-| Tempo de geracao | 6.83s |
-| Duracao do audio | 7.59s |
-| **RTF (Real-Time Factor)** | **0.899** |
-| Velocidade | 1.1x tempo real |
-| Memoria GPU usada | ~6.2 GiB |
-
-## Conclusao
-
-O Orpheus TTS na RTX 4090 consegue rodar **mais rapido que tempo real** (RTF < 1.0).
-
-## Arquivos
-
-- `test_orpheus.py` - Script de teste
-- `test_orpheus_output.wav` - Audio gerado (24kHz, mono, 16-bit)
-
-## Texto do Teste
-
-> "Hello! This is a test of the Orpheus text to speech system running on an RTX 4090. How does it sound?"
-
-## Proximos Passos - Otimizacoes
-
-Para suportar 10 usuarios simultaneos:
-
-1. **FP8 quantization** - 1.5-1.6x speedup
-2. **Continuous batching** - Escala throughput
-3. **TensorRT-LLM** - 1.3-1.5x speedup
-4. **Speculative decoding** - 1.5-2x speedup
-
-Estimativa com otimizacoes: RTF ~0.3-0.4, suportando 8-12 streams simultaneos.

orpheus-tts/2024-12-24/test_orpheus.py

@@ -1,54 +0,0 @@
-import time
-import wave
-
-def main():
-    from orpheus_tts import OrpheusModel
-
-    print("="*50)
-    print("TESTE ORPHEUS TTS - RTX 4090")
-    print("="*50)
-
-    print("\n[1] Carregando modelo Orpheus TTS...")
-    start_load = time.time()
-    model = OrpheusModel(model_name="canopylabs/orpheus-3b-0.1-ft")
-    load_time = time.time() - start_load
-    print(f"    Modelo carregado em {load_time:.2f}s")
-
-    text = "Hello! This is a test of the Orpheus text to speech system running on an RTX 4090. How does it sound?"
-    print(f"\n[2] Gerando audio para:")
-    print(f'    "{text}"')
-
-    start_gen = time.time()
-    audio_chunks = []
-    # Use prompt= instead of text=
-    for chunk in model.generate_speech(prompt=text, voice="tara"):
-        audio_chunks.append(chunk)
-    gen_time = time.time() - start_gen
-
-    audio_data = b"".join(audio_chunks)
-
-    with wave.open("/root/test_output.wav", "wb") as wf:
-        wf.setnchannels(1)
-        wf.setsampwidth(2)
-        wf.setframerate(24000)
-        wf.writeframes(audio_data)
-
-    audio_duration = len(audio_data) / (24000 * 2)
-    rtf = gen_time / audio_duration
-
-    print(f"\n[3] === RESULTADOS ===")
-    print(f"    Tempo de geracao: {gen_time:.2f}s")
-    print(f"    Duracao do audio: {audio_duration:.2f}s")
-    print(f"    RTF (Real-Time Factor): {rtf:.3f}")
-    print(f"    Velocidade: {1/rtf:.1f}x tempo real")
-    print(f"    Audio salvo em: /root/test_output.wav")
-
-    if rtf < 1.0:
-        print(f"\n    [OK] SUCESSO! Roda mais rapido que tempo real!")
-    else:
-        print(f"\n    [X] Nao consegue tempo real (precisa RTF < 1.0)")
-
-    print("="*50)
-
-if __name__ == "__main__":
-    main()

orpheus-tts/2024-12-24/test_orpheus_output.wav

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:63d9c7444bcf62b2346ac429111e8d74b1ca1cad516ecdbdad15d6b6b25701a3
-size 364588