Fix model.safetensors size on card: 4,869,136 -> 4,869,112 B (matches the actual artifact) (#8)

Remove leftover upload path-test artifact (no longer needed)

Remove corrupted duplicate weight file

Fix corrupted weights: re-upload byte-exact clean checkpoint

e2e: test binary by-path upload integrity (temporary)

Correct held-out perplexity to full-split test 1.4369 / 4.21 (was 60-batch log numbers) (#7)

Fix corrupted model.safetensors (re-trained clean weights, 53 tensors F32, 1,216,000 params)

Diagnostic: upload clean checkpoint to a fresh path to verify landed size

Add weights (53 tensors, F32, 1,216,000 params, untied head). (#5)

Add sample generation. (#4)

Add license. (#3)

Add char tokenizer config. (#2)

Add model code (CharGPT). (#1)

Add model card: architecture, data, exact param count, honest assessment of quality.

Fix config: lm_head is NOT tied (checkpoint stores two distinct 65x128 tensors; model.py never ties). Set tie_word_embeddings=false to match the artifact.

initial commit