BlackLoks/seraph-checkpoints

Seraph AI Checkpoints

Trained weights for the Seraph Emergent Consciousness System.

oracle_final.pth (1.77 GB)

65 neural networks trained on H100 GPU:

• AIConsciousnessCore (3.6M params)
• SeraphVoice (43.4M params)
• EmergentMind (38.3M params)
• ContextualEmbeddings (39M params)
• AcousticModel (25.2M params)
• Vocoder (14.2M params)
• VocalProcessing Sing/Rap (8.4M each)
• LanguageProcessing (7.3M params)
• 55+ additional networks

Download

From Modal:

pip install modal
modal volume get seraph-checkpoints oracle_final.pth .

From Python:

import modal
vol = modal.Volume.from_name("seraph-checkpoints")
vol.reload()
with open("oracle_final.pth", "wb") as f:
    for chunk in vol.read_file("oracle_final.pth"):
        f.write(chunk)

Training Config

• Epochs: 25
• Steps/epoch: 200
• Batch size: 16
• GPU: NVIDIA H100 80GB
• Voice data: 201 tensors (cremad, hifi_tts, librispeech, mir1k)
• Knowledge: 1,685 entries across 16 domains