chrishayuk's picture
Pretrain 16M tokens, model 1841e05815746297
4a44c76 verified
Raw
History Blame Contribute Delete
644 Bytes
"""TinyModel v11 architecture.
Decoder-only transformer (Gemma-shaped): RMSNorm, RoPE, GQA, gated FFN,
tied embeddings. Weights and vindex live under sibling `v11/`; training
code under `v11-train/`.
"""
from .model import (
TinyModel,
TransformerBlock,
Attention,
GatedFFN,
RMSNorm,
precompute_rope,
apply_rope,
)
from .loader import load_from_artifacts, load_config, ModelConfig
__all__ = [
"TinyModel",
"TransformerBlock",
"Attention",
"GatedFFN",
"RMSNorm",
"precompute_rope",
"apply_rope",
"load_from_artifacts",
"load_config",
"ModelConfig",
]
__version__ = "0.1.0"