ttllm-nano (+ v2 BPE)

The shape of total transparency at nano scale — not a frontier model, not soft tissue cosplay of one.

Trained on this Mac (Apple M1 / MPS) with a character-level Transformer on Project Gutenberg public-domain train slices (Pride and Prejudice, Alice, Frankenstein). Full downloads retained and hashed. Dense intermediate checkpoints, metrics JSONL, Merkle release manifest, ttlink over train docs, hash-chained stream.

models/ttllm-nano/ on GitHub
Data card · Transparency scorecard
Nano public stream JSON
Nano ttlink index
models/ttllm-nano-v2/ — BPE + denser model + ppl metrics

Reproduce

cd daisychainrrr
python3 models/ttllm-nano/code/prepare_data.py
python3 models/ttllm-nano/code/train.py --steps 800
python3 models/ttllm-nano/code/seal_release.py
python3 models/ttllm-nano/code/generate.py --prompt "Alice "

Tombstones (required honesty)

Not multi-trillion-token data · not 32B · not production FM-index
Sample text quality is expected to be poor; the product under test is the process skeleton