File size: 1,140 Bytes
a4c1304 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 | {
"model_type": "rodan-modern",
"architecture": "ModernLM",
"framework": "mlx",
"stage": "chat",
"base_model": "Rodan-10M (v9, PLE-free)",
"params": 10410000,
"vocab_size": 8194,
"dim": 320,
"n_layers": 8,
"n_heads": 8,
"n_kv_heads": 1,
"head_dim": 40,
"ffn_hidden": 768,
"max_len": 512,
"rope_base": 200000.0,
"norm": "rmsnorm",
"norm_eps": 1e-5,
"activation": "swiglu",
"qk_norm": true,
"tied_embeddings": true,
"value_residual": true,
"ple_rank": 0,
"lrm": true,
"attention": "mqa",
"chat_template": "chatml",
"chat_tokens": {"im_start": 8192, "im_end": 8193},
"eot_id": 0,
"tokenizer": "byte-level BPE (8k) + 2 ChatML specials = 8194",
"recommended_decode": "greedy + repetition_penalty 1.3 + no-repeat-3gram (tiny models loop under pure greedy)",
"notes": "Warm-started from Rodan-10M v9 (PLE-free). Instruction fold: smol-smoltalk ChatML + 45% curated replay (continued-pretrain, not masked SFT). Load with model_opt.ModernLM(ModernConfig(**fields)) + load_weights('model.safetensors'). Wrap prompts in ChatML: <|im_start|>user\\n{q}<|im_end|>\\n<|im_start|>assistant\\n"
}
|