| from transformers import PretrainedConfig | |
| class PulvisConfig(PretrainedConfig): | |
| model_type = "pulvis" | |
| def __init__(self, vocab_size=4096, hidden_size=192, num_attention_heads=6, num_key_value_heads=2, head_dim=32, | |
| intermediate_size=368, prelude_layers=1, core_layers=3, core_loops=3, coda_layers=3, | |
| max_position_embeddings=1024, rope_theta=10000.0, rms_norm_eps=1e-6, logit_cap=15.0, | |
| bos_token_id=0, eos_token_id=0, pad_token_id=1, tie_word_embeddings=False, **kwargs): | |
| self.vocab_size = vocab_size | |
| self.hidden_size = hidden_size | |
| self.num_attention_heads = num_attention_heads | |
| self.num_key_value_heads = num_key_value_heads | |
| self.head_dim = head_dim | |
| self.intermediate_size = intermediate_size | |
| self.prelude_layers = prelude_layers | |
| self.core_layers = core_layers | |
| self.core_loops = core_loops | |
| self.coda_layers = coda_layers | |
| self.num_hidden_layers = prelude_layers + core_layers * core_loops + coda_layers | |
| self.max_position_embeddings = max_position_embeddings | |
| self.rope_theta = rope_theta | |
| self.rms_norm_eps = rms_norm_eps | |
| self.logit_cap = logit_cap | |
| super().__init__(bos_token_id=bos_token_id, eos_token_id=eos_token_id, pad_token_id=pad_token_id, | |
| tie_word_embeddings=tie_word_embeddings, **kwargs) | |