Microreason-43M / extra-sft /config.json
arpjw's picture
MicroReason-43M: sft-long, grpo and extra-sft checkpoints with model card
5352e16 verified
Raw History Blame Contribute Delete
647 Bytes
{
"architectures": [
"MicroReasonForCausalLM"
],
"bos_token_id": 1,
"d_model": 512,
"dtype": "float32",
"eos_token_id": 2,
"head_dim": 64,
"init_std": 0.02,
"max_seq_len": 2048,
"mlp_hidden": 1408,
"model_type": "microreason",
"n_heads": 8,
"n_kv_heads": 8,
"n_layers": 12,
"name": "microreason-43m",
"norm_eps": 1e-05,
"pad_token_id": 0,
"rope_theta": 10000.0,
"tie_word_embeddings": false,
"transformers_version": "5.22.2",
"vocab_size": 8192,
"auto_map": {
"AutoConfig": "modeling_microreason.MicroReasonConfig",
"AutoModelForCausalLM": "modeling_microreason.MicroReasonForCausalLM"
}
}