File size: 2,050 Bytes
fd1a388 | 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 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 | # Copyright (C) Michael Lee (李登淳) 2026. All rights reserved.
# Open-source under the MIT License. See LICENSE for details.
from transformers import PretrainedConfig
class TinyMixtralConfig(PretrainedConfig):
model_type = "tinymixtral"
def __init__(
self,
vocab_size: int = 32000,
hidden_size: int = 896,
num_hidden_layers: int = 10,
num_attention_heads: int = 16,
num_key_value_heads: int = 4,
head_dim: int = 56,
max_position_embeddings: int = 2048,
num_shared_experts: int = 1,
num_routed_experts: int = 6,
num_experts_per_tok: int = 2,
expert_intermediate_size: int = 2389,
router_aux_loss_coef: float = 0.01,
router_jitter_noise: float = 0.01,
rms_norm_eps: float = 1e-6,
rope_theta: float = 1_000_000.0,
attention_dropout: float = 0.0,
tie_word_embeddings: bool = True,
initializer_range: float = 0.02,
**kwargs,
):
super().__init__(tie_word_embeddings=tie_word_embeddings, **kwargs)
self.vocab_size = vocab_size
self.hidden_size = hidden_size
self.num_hidden_layers = num_hidden_layers
self.num_attention_heads = num_attention_heads
self.num_key_value_heads = num_key_value_heads
self.head_dim = head_dim
self.max_position_embeddings = max_position_embeddings
self.num_shared_experts = num_shared_experts
self.num_routed_experts = num_routed_experts
self.num_experts_per_tok = num_experts_per_tok
self.expert_intermediate_size = expert_intermediate_size
self.router_aux_loss_coef = router_aux_loss_coef
self.router_jitter_noise = router_jitter_noise
self.rms_norm_eps = rms_norm_eps
self.rope_theta = rope_theta
self.attention_dropout = attention_dropout
self.initializer_range = initializer_range
@property
def num_local_experts(self) -> int:
return self.num_shared_experts + self.num_routed_experts
|