mlnomad/yatnmn-softplus-ca-d12-chinchilla-261M-pytorch
Fix: lazy non-persistent RoPE buffers — HF from_pretrained's meta-init was leaving persistent rope_cos/rope_sin uninitialised, producing NaN logits. Buffers are now computed on first forward.
Fix: lazy non-persistent RoPE buffers — HF from_pretrained's meta-init was leaving persistent rope_cos/rope_sin uninitialised, producing NaN logits. Buffers are now computed on first forward.
Fix: lazy non-persistent RoPE buffers — HF from_pretrained's meta-init was leaving persistent rope_cos/rope_sin uninitialised, producing NaN logits. Buffers are now computed on first forward.
Fix: lazy non-persistent RoPE buffers — HF from_pretrained's meta-init was leaving persistent rope_cos/rope_sin uninitialised, producing NaN logits. Buffers are now computed on first forward.
Fix: lazy non-persistent RoPE buffers — HF from_pretrained's meta-init was leaving persistent rope_cos/rope_sin uninitialised, producing NaN logits. Buffers are now computed on first forward.
Fix: lazy non-persistent RoPE buffers — HF from_pretrained's meta-init was leaving persistent rope_cos/rope_sin uninitialised, producing NaN logits. Buffers are now computed on first forward.
Fix: lazy non-persistent RoPE buffers — HF from_pretrained's meta-init was leaving persistent rope_cos/rope_sin uninitialised, producing NaN logits. Buffers are now computed on first forward.
Fix: lazy non-persistent RoPE buffers — HF from_pretrained's meta-init was leaving persistent rope_cos/rope_sin uninitialised, producing NaN logits. Buffers are now computed on first forward.
Fix: lazy non-persistent RoPE buffers — HF from_pretrained's meta-init was leaving persistent rope_cos/rope_sin uninitialised, producing NaN logits. Buffers are now computed on first forward.
Upload yatnmn_gpt.py with huggingface_hub
yatnmn-softplus-ca-d12-chinchilla-261M-pytorch — constant alpha ablation
initial commit
