Trained model checkpoints for the nanochat language model with a frozen WASM coprocessor. Latest Run Date : 2026-03-22 16:00 Model : full d34 (d=2176, 34L, 17H) Params : 4,686,156,660 trainable, 368,804 frozen Device : cuda Total time : 3650.0s (60.8 min) Results SFT: 1 epochs, loss 4.0005 - 4.0005 Repo Structure latest/ -- most recent checkpoints (for resuming training) runs/ / -- per-run...
Model source
Source description
Sources & provenance
1 active sourcehighest_eligible_score
Verified Jul 11
Source evidence
3 excerptsTrained model checkpoints for the nanochat language model with a frozen WASM coprocessor. Latest Run Date : 2026-03-22 16:00 Model : full d34 (d=2176, 34L, 17H) Params : 4,686,156,660 trainable, 368,804 frozen Device : cuda Total time : 3650.0s (60.8 min)…
tags: nanochat wasm-coprocessor pytorch license: mit
eastlondoner/nanochat-wasm-coprocessor