from collections import OrderedDict
from typing import Any, List, Mapping, Optional

from transformers import PreTrainedTokenizer, TensorType, is_torch_available
from transformers.configuration_utils import PretrainedConfig
from transformers.utils import logging

logger = logging.get_logger(__name__)


class NanoConfig(PretrainedConfig):
    model_type = "nano"
    keys_to_ignore_at_inference = ["past_key_values"]
    attribute_map = {
        "hidden_size": "hidden_size",
        "max_position_embeddings": "max_position_embeddings",
        "num_attention_heads": "num_attention_heads",
        "num_hidden_layers": "num_hidden_layers",
    }

    def __init__(
        self,
        vocab_size=32000,
        max_position_embeddings=2048,
        expanded_wte_size=None,
        expanded_lm_head_size=None,
        hidden_size=768,
        kv_hidden_size=None,  # in case you want to use cross-attention
        num_hidden_layers=10,
        num_attention_heads=12,
        intermediate_size=None,
        activation_function="silu",
        layer_norm_epsilon=1e-6,
        initializer_range=0.02,
        use_cache=True,
        bos_token_id=1,
        eos_token_id=2,
        combined_qkv=True,
        use_bias=False,
        lm_head_projection_bias=False,
        lm_head_bias=False,
        layernorm="llamarmsnorm", # layernorm, llamarmsnorm
        rope_scaling=None,
        rope_theta=10000,
        ffn="llama-like",
        experimental_full_adaption_rank = None, # 8
        full_adaptation_has_pre_proj = True,
        pre_proj_dim = 1536,
        full_adaptation_type="no", # "lora", "no", "linear", "linear-r", "linear-ra"
        tie_word_embeddings=False,
        residual_alpha=False,
        **kwargs,
    ):
        self.residual_alpha = residual_alpha
        self.pre_proj_dim = pre_proj_dim
        self.full_adaptation_has_pre_proj = full_adaptation_has_pre_proj
        self.full_adaptation_type = full_adaptation_type
        self.tie_word_embeddings = tie_word_embeddings
        self.experimental_full_adaption_rank = experimental_full_adaption_rank
        self.ffn = ffn
        self.rope_theta=rope_theta
        self.layernorm = layernorm
        self.rope_scaling=rope_scaling
        self.lm_head_projection_bias = lm_head_projection_bias
        self.kv_hidden_size = kv_hidden_size
        self.lm_head_bias = lm_head_bias
        self.use_bias = use_bias
        self.expanded_wte_size = expanded_wte_size
        self.expanded_lm_head_size = expanded_lm_head_size
        self.combined_qkv = combined_qkv
        self.vocab_size = vocab_size
        self.max_position_embeddings = max_position_embeddings
        self.hidden_size = hidden_size
        self.num_hidden_layers = num_hidden_layers
        self.num_attention_heads = num_attention_heads
        self.intermediate_size = (
            intermediate_size if intermediate_size is not None else hidden_size * 4
        )
        self.activation_function = activation_function
        self.layer_norm_epsilon = layer_norm_epsilon
        self.initializer_range = initializer_range

        self.use_cache = use_cache

        self.bos_token_id = bos_token_id
        self.eos_token_id = eos_token_id

        super().__init__(bos_token_id=bos_token_id, eos_token_id=eos_token_id, **kwargs)