config.json

{
  "architectures": [
    "TAMELM"
  ],
  "attention_backend": "flex",
  "attention_top_k": 2,
  "auto_map": {
    "AutoConfig": "modeling_tamelm.TAMEConfig",
    "AutoModelForCausalLM": "modeling_tamelm.TAMELM"
  },
  "balance_coef": 0.01,
  "bos_token_id": 1,
  "depth_dim": 256,
  "depth_gate_init": 0.01,
  "depth_heads": 4,
  "depth_mode": "attention",
  "depth_slots": 4,
  "dtype": "float32",
  "embed_dim": 768,
  "eos_token_id": 2,
  "expert_backend": "grouped",
  "expert_hidden_dim": 3072,
  "expert_top_k": 2,
  "local_window": 256,
  "loss_chunk_size": 256,
  "max_seq_len": 2048,
  "model_type": "tamelm_two_axis",
  "n_kv_heads": 4,
  "n_q_heads": 32,
  "num_experts": 8,
  "num_layers": 12,
  "pad_token_id": 0,
  "rope_type": "blackhole",
  "strided_every": 4,
  "strided_recent": 128,
  "tie_word_embeddings": true,
  "tile": 128,
  "tokenizer_sha256": "a4ff24e4d2930c8b41c73d29e94474d4ce4db73f38f9891ea152d43981dcf3e5",
  "transformers_version": "5.5.0",
  "use_cache": true,
  "vocab_size": 44469
}