model.py

# Copyright © 2023-2024 Apple Inc.

from dataclasses import dataclass
from typing import Any, Dict, List, Optional, Union

import mlx.core as mx
import mlx.nn as nn
from mlx.nn.layers.distributed import shard_linear

from mlx_lm.models.activations import swiglu
from mlx_lm.models.base import BaseModelArgs, create_attention_mask, scaled_dot_product_attention
from mlx_lm.models.cache import KVCache, RotatingKVCache
from mlx_lm.models.rope_utils import initialize_rope


@dataclass
class ModelArgs(BaseModelArgs):
    model_type: str
    hidden_size: int
    num_hidden_layers: int
    intermediate_size: int
    num_attention_heads: int
    rms_norm_eps: float
    vocab_size: int
    head_dim: Optional[int] = None
    max_position_embeddings: Optional[int] = None
    num_key_value_heads: Optional[int] = None
    attention_bias: bool = False
    mlp_bias: bool = False
    rope_theta: float = 10000
    rope_traditional: bool = False
    rope_scaling: Optional[Dict[str, Union[float, str]]] = None
    rope_parameters: Optional[Dict[str, Union[float, str]]] = None
    tie_word_embeddings: bool = True
    layernorm_num_groups: int = 1
    layer_types: Optional[List[str]] = None
    sliding_window: Optional[int] = None

    def __post_init__(self):
        if self.num_key_value_heads is None:
            self.num_key_value_heads = self.num_attention_heads

        if self.layer_types is None:
            self.layer_types = ["full_attention"] * self.num_hidden_layers

        # K2-Horizon stores RoPE settings in `rope_parameters`.
        if self.rope_parameters is not None:
            self.rope_theta = float(
                self.rope_parameters.get("rope_theta", self.rope_theta)
            )


class GroupRMSNorm(nn.Module):
    """K2-Horizon's grouped RMS normalization (T5LayerNorm equivalent)."""

    def __init__(self, dims: int, eps: float = 1e-5, groups: int = 1):
        super().__init__()
        if dims % groups != 0:
            raise ValueError(f"hidden size {dims} must be divisible by groups {groups}")
        self.weight = mx.ones((dims,))
        self.groups = groups
        self.eps = eps

    def __call__(self, x: mx.array) -> mx.array:
        shape = x.shape
        x = mx.reshape(x, (*shape[:-1], self.groups, -1))
        x = mx.fast.rms_norm(x, weight=None, eps=self.eps)
        return self.weight * mx.reshape(x, shape)


class Attention(nn.Module):
    def __init__(self, args: ModelArgs):
        super().__init__()

        dim = args.hidden_size
        self.n_heads = n_heads = args.num_attention_heads
        self.n_kv_heads = n_kv_heads = args.num_key_value_heads

        self.head_dim = head_dim = args.head_dim or args.hidden_size // n_heads

        self.scale = head_dim**-0.5
        if hasattr(args, "attention_bias"):
            attention_bias = args.attention_bias
        else:
            attention_bias = False

        self.q_proj = nn.Linear(dim, n_heads * head_dim, bias=attention_bias)
        self.k_proj = nn.Linear(dim, n_kv_heads * head_dim, bias=attention_bias)
        self.v_proj = nn.Linear(dim, n_kv_heads * head_dim, bias=attention_bias)
        self.o_proj = nn.Linear(n_heads * head_dim, dim, bias=attention_bias)

        self.rope = initialize_rope(
            self.head_dim,
            args.rope_theta,
            args.rope_traditional,
            args.rope_scaling,
            args.max_position_embeddings,
        )

    def __call__(
        self,
        x: mx.array,
        mask: Optional[mx.array] = None,
        cache: Optional[Any] = None,
    ) -> mx.array:
        B, L, D = x.shape

        queries, keys, values = self.q_proj(x), self.k_proj(x), self.v_proj(x)

        # Prepare the queries, keys and values for the attention computation
        queries = queries.reshape(B, L, self.n_heads, -1).transpose(0, 2, 1, 3)
        keys = keys.reshape(B, L, self.n_kv_heads, -1).transpose(0, 2, 1, 3)
        values = values.reshape(B, L, self.n_kv_heads, -1).transpose(0, 2, 1, 3)

        if cache is not None:
            queries = self.rope(queries, offset=cache.offset)
            keys = self.rope(keys, offset=cache.offset)
            keys, values = cache.update_and_fetch(keys, values)
        else:
            queries = self.rope(queries)
            keys = self.rope(keys)

        output = scaled_dot_product_attention(
            queries, keys, values, cache=cache, scale=self.scale, mask=mask
        )

        output = output.transpose(0, 2, 1, 3).reshape(B, L, -1)
        return self.o_proj(output)


class MLP(nn.Module):
    def __init__(self, args: ModelArgs):
        super().__init__()

        dim = args.hidden_size
        hidden_dim = args.intermediate_size
        if hasattr(args, "mlp_bias"):
            mlp_bias = args.mlp_bias
        else:
            mlp_bias = False

        self.gate_proj = nn.Linear(dim, hidden_dim, bias=mlp_bias)
        self.down_proj = nn.Linear(hidden_dim, dim, bias=mlp_bias)
        self.up_proj = nn.Linear(dim, hidden_dim, bias=mlp_bias)

    def __call__(self, x) -> mx.array:
        return self.down_proj(swiglu(self.gate_proj(x), self.up_proj(x)))


class TransformerBlock(nn.Module):
    def __init__(self, args: ModelArgs, use_sliding: bool = False):
        super().__init__()
        self.num_attention_heads = args.num_attention_heads
        self.hidden_size = args.hidden_size
        self.use_sliding = use_sliding
        self.self_attn = Attention(args)
        self.mlp = MLP(args)
        self.input_layernorm = GroupRMSNorm(
            args.hidden_size,
            eps=args.rms_norm_eps,
            groups=args.layernorm_num_groups,
        )
        self.post_attention_layernorm = GroupRMSNorm(
            args.hidden_size,
            eps=args.rms_norm_eps,
            groups=args.layernorm_num_groups,
        )
        self.args = args

    def __call__(
        self,
        x: mx.array,
        mask: Optional[mx.array] = None,
        cache: Optional[Any] = None,
    ) -> mx.array:
        r = self.self_attn(self.input_layernorm(x), mask, cache)
        h = x + r
        r = self.mlp(self.post_attention_layernorm(h))
        out = h + r
        return out


class LlamaModel(nn.Module):
    def __init__(self, args: ModelArgs):
        super().__init__()
        self.args = args
        self.vocab_size = args.vocab_size
        self.num_hidden_layers = args.num_hidden_layers
        self.layer_types = args.layer_types
        self.sliding_window = args.sliding_window
        assert self.vocab_size > 0
        self.embed_tokens = nn.Embedding(args.vocab_size, args.hidden_size)
        self.layers = [
            TransformerBlock(args=args, use_sliding=layer_type == "sliding_attention")
            for layer_type in self.layer_types
        ]
        self.norm = GroupRMSNorm(
            args.hidden_size,
            eps=args.rms_norm_eps,
            groups=args.layernorm_num_groups,
        )
        self.fa_idx = self.layer_types.index("full_attention")
        self.swa_idx = None
        for e, l in enumerate(self.layers):
            if l.use_sliding:
                self.swa_idx = e
                break

    def __call__(
        self,
        inputs: mx.array,
        cache=None,
        input_embeddings: Optional[mx.array] = None,
    ):
        if input_embeddings is not None:
            h = input_embeddings
        else:
            h = self.embed_tokens(inputs)

        if cache is None:
            cache = [None] * len(self.layers)

        fa_mask = create_attention_mask(h, cache[self.fa_idx])
        if self.swa_idx is not None:
            swa_mask = create_attention_mask(
                h, cache[self.swa_idx], window_size=self.sliding_window
            )

        for layer, cache in zip(self.layers, cache):
            mask = swa_mask if layer.use_sliding else fa_mask
            h = layer(h, mask, cache=cache)

        return self.norm(h)


class Model(nn.Module):
    def __init__(self, args: ModelArgs):
        super().__init__()
        self.args = args
        self.model_type = args.model_type
        self.model = LlamaModel(args)
        if not args.tie_word_embeddings:
            self.lm_head = nn.Linear(args.hidden_size, args.vocab_size, bias=False)

    def __call__(
        self,
        inputs: mx.array,
        cache=None,
        input_embeddings: Optional[mx.array] = None,
    ):
        out = self.model(inputs, cache, input_embeddings)
        if self.args.tie_word_embeddings:
            out = self.model.embed_tokens.as_linear(out)
        else:
            out = self.lm_head(out)
        return out

    def sanitize(self, weights):
        # Remove unused precomputed rotary freqs
        weights = {
            k: v for k, v in weights.items() if "self_attn.rotary_emb.inv_freq" not in k
        }
        if self.args.tie_word_embeddings:
            weights.pop("lm_head.weight", None)
        return weights

    def shard(self, group: Optional[mx.distributed.Group] = None):
        group = group or mx.distributed.init()
        N = group.size()
        for layer in self.model.layers:
            # Shard the self attention
            layer.self_attn.q_proj = shard_linear(
                layer.self_attn.q_proj, "all-to-sharded", group=group
            )
            layer.self_attn.k_proj = shard_linear(
                layer.self_attn.k_proj, "all-to-sharded", group=group
            )
            layer.self_attn.v_proj = shard_linear(
                layer.self_attn.v_proj, "all-to-sharded", group=group
            )
            layer.self_attn.o_proj = shard_linear(
                layer.self_attn.o_proj, "sharded-to-all", group=group
            )
            layer.self_attn.n_heads //= N
            layer.self_attn.n_kv_heads //= N

            # Shard the MLP
            layer.mlp.gate_proj = shard_linear(
                layer.mlp.gate_proj, "all-to-sharded", group=group
            )
            layer.mlp.down_proj = shard_linear(
                layer.mlp.down_proj, "sharded-to-all", group=group
            )
            layer.mlp.up_proj = shard_linear(
                layer.mlp.up_proj, "all-to-sharded", group=group
            )

    @property
    def layers(self):
        return self.model.layers

    def make_cache(self):
        return [
            (
                RotatingKVCache(max_size=self.model.sliding_window)
                if layer.use_sliding
                else KVCache()
            )
            for layer in self.layers
        ]