CMA-20M / config.json
User01110's picture
Evaluated step 2000: WikiText BPB 1.3592
6e07d52 verified
Raw
History Blame Contribute Delete
888 Bytes
{
"architectures": [
"CMAForCausalLM"
],
"auto_map": {
"AutoConfig": "modeling_cma.CMAConfig",
"AutoModelForCausalLM": "modeling_cma.CMAForCausalLM"
},
"model_type": "cma",
"vocab_size": 8192,
"seq_len": 1024,
"max_position_embeddings": 1024,
"n_positions": 1024,
"n_ctx": 1024,
"d_model": 336,
"n_layers": 13,
"n_heads": 6,
"n_kv_heads": 2,
"chunk": 24,
"cma_heads": 3,
"expand": 3,
"cma_identity_prob": 0.9,
"route_mode": "full_attention",
"route_head_shifts": [
-1,
0,
1
],
"tokenizer_source": "BananaMind/BananaMind-2-Mini",
"tokenizer_revision": "84a0afb98db902caf07a1e949676d9fef9e5cf9e",
"bos_token_id": 1,
"eos_token_id": 2,
"pad_token_id": 0,
"tie_word_embeddings": true,
"safe_serialization": true,
"minimum_torch_version": "2.5",
"torch_dtype": "bfloat16",
"transformers_version": "5.13.0"
}