Sep 27 13:59 Fix card: replace hand-written sample sentences with real output from the shipped weights, and correct the capability claims to match what the model actually produces 13:46 Add model card (honest: arch, data, measured val PPL 48.03, 0/15 degenerate) (#6 ) - Add model card (honest: arch, data, measured val PPL 48.03, 0/15 degenerate) (a7cb36a13479a328e836a34cb2b6df8e2baaab50) 13:46 Add config.json (architecture params, verified 6,162,688 params) (#5 ) - Add config.json (architecture params, verified 6,162,688 params) (1bff3cbee100a5e8b443f50d7e31393041f0e693) 13:46 Add exact eval script (val PPL + generation + degeneracy check) (#4 ) - Add exact eval script (val PPL + generation + degeneracy check) (4ba2a1c6e79f3b8881cde21d2c6ae2601bea3626) 13:46 Add exact training script (defines CompactLM class) (#3 ) - Add exact training script (defines CompactLM class) (a4bf8f170ad7f00fd0efb827f7ef1161e02585ff) 13:46 Add gollem_eval byte-level BPE tokenizer (12,288 vocab) (#2 ) - Add gollem_eval byte-level BPE tokenizer (12,288 vocab) (1b61941a175dac0067c59522a9da24937b6efcd0) 13:46 Add CompactLM-5M weights (39 tensors, F32, 6,162,688 params, tied embeddings) (#1 ) - Add CompactLM-5M weights (39 tensors, F32, 6,162,688 params, tied embeddings) (07721325cd8e52d0717b63cc3fbafa6129df4a91) 12:51 Add fresh eval results (val ppl + 15 samples + degeneracy check) 12:51 Add CompactLM-5M checkpoint (6,162,688 params, 39 tensors) 12:51 Add CompactLM-5M: from-scratch LLaMA-style 6.2M-param English LM on fineweb-edu 12:51 Add architecture config 12:50 initial commit