46da74fea2
* Unify attention mask creation in LLMs.
Currently, each model implementation in `mlx-examples/llms/models` has ad-hoc
code to create a mask for the attention mechanism. This usually takes the form:
```
mask = None
if h.shape[1] > 1:
mask = nn.MultiHeadAttention.create_additive_causal_mask(h.shape[1])
mask = mask.astype(h.dtype)
```
This correctly creates a mask only if the input consists of more than one token.
But this code assumes the multi-token input is at the beginning of inference.
If, for example, we are evaluating multiple tokens because of speculative
decoding or prompt cache reuse, this mask will not have the correct shape and
and will cause the raising of an exception in the attention computation.
Some of the models correctly implement the mask creation with code like this:
```
mask = None
if h.shape[1] > 1:
mask = create_additive_causal_mask(
h.shape[1], cache[0].offset if cache is not None else 0
)
mask = mask.astype(h.dtype)
```
This commit unifies the attention mask creation for all models with a new
function `create_attention_mask`, reducing code duplication and helping all
models support inference performance enhancements like those mentioned above.
* Allow batches in LLM key-value cache
The current implementation of the LLM key-value cache assumes that
the input batch is of size 1. Input batching (evaluating multiple
alterative inputs at the same time) can be a valuable tool for
speculative sampling and other techniques.
This change removes the hard-coded batch size from the code that
resizes the key-value cache.
* Simplify causal mask creation
Use the same codepath regardless of whether there's an offset or
not. Addresses [this comment](https://github.com/ml-explore/mlx-examples/pull/911#discussion_r1691459717).
* Use old-style type annotation to avoid linter error
216 lines
6.8 KiB
Python
216 lines
6.8 KiB
Python
import math
|
|
from dataclasses import dataclass
|
|
from typing import Tuple
|
|
|
|
import mlx.core as mx
|
|
import mlx.nn as nn
|
|
|
|
from .base import BaseModelArgs, create_attention_mask
|
|
|
|
|
|
@dataclass
|
|
class ModelArgs(BaseModelArgs):
|
|
model_type: str
|
|
vocab_size: int
|
|
hidden_size: int
|
|
num_attention_heads: int
|
|
num_hidden_layers: int
|
|
num_key_value_heads: int
|
|
intermediate_size: int
|
|
rope_theta: float
|
|
use_qkv_bias: bool
|
|
partial_rotary_factor: float
|
|
layer_norm_eps: float
|
|
use_parallel_residual: bool = False
|
|
qk_layernorm: bool = False
|
|
|
|
|
|
class LayerNormPerHead(nn.Module):
|
|
|
|
def __init__(self, head_dim, num_heads, eps):
|
|
super().__init__()
|
|
self.norms = [
|
|
nn.LayerNorm(head_dim, eps=eps, bias=False) for _ in range(num_heads)
|
|
]
|
|
self.eps = eps
|
|
|
|
def __call__(self, x):
|
|
w = mx.stack([n.weight for n in self.norms])
|
|
return w * mx.fast.layer_norm(x, None, None, self.eps)
|
|
|
|
|
|
class Attention(nn.Module):
|
|
def __init__(self, config: ModelArgs):
|
|
super().__init__()
|
|
|
|
self.hidden_size = config.hidden_size
|
|
self.num_heads = config.num_attention_heads
|
|
self.head_dim = self.hidden_size // self.num_heads
|
|
self.num_key_value_heads = config.num_key_value_heads
|
|
self.rope_theta = config.rope_theta
|
|
self.partial_rotary_factor = config.partial_rotary_factor
|
|
|
|
if (self.head_dim * self.num_heads) != self.hidden_size:
|
|
raise ValueError(
|
|
f"hidden_size must be divisible by num_heads (got `hidden_size`: {self.hidden_size}"
|
|
f" and `num_heads`: {self.num_heads})."
|
|
)
|
|
|
|
self.q_proj = nn.Linear(
|
|
self.hidden_size, self.num_heads * self.head_dim, bias=config.use_qkv_bias
|
|
)
|
|
self.k_proj = nn.Linear(
|
|
self.hidden_size,
|
|
self.num_key_value_heads * self.head_dim,
|
|
bias=config.use_qkv_bias,
|
|
)
|
|
self.v_proj = nn.Linear(
|
|
self.hidden_size,
|
|
self.num_key_value_heads * self.head_dim,
|
|
bias=config.use_qkv_bias,
|
|
)
|
|
self.o_proj = nn.Linear(
|
|
self.num_heads * self.head_dim, self.hidden_size, bias=False
|
|
)
|
|
|
|
self.rope = nn.RoPE(
|
|
int(self.partial_rotary_factor * self.head_dim),
|
|
traditional=False,
|
|
base=self.rope_theta,
|
|
)
|
|
|
|
self.qk_layernorm = config.qk_layernorm
|
|
if self.qk_layernorm:
|
|
self.q_layernorm = LayerNormPerHead(
|
|
self.head_dim, self.num_heads, eps=config.layer_norm_eps
|
|
)
|
|
self.k_layernorm = LayerNormPerHead(
|
|
self.head_dim, self.num_key_value_heads, eps=config.layer_norm_eps
|
|
)
|
|
|
|
def __call__(self, x, mask=None, cache=None):
|
|
queries, keys, values = self.q_proj(x), self.k_proj(x), self.v_proj(x)
|
|
|
|
# Extract some shapes
|
|
B, L, D = queries.shape
|
|
|
|
queries = queries.reshape(B, L, self.num_heads, -1)
|
|
keys = keys.reshape(B, L, self.num_key_value_heads, -1)
|
|
if self.qk_layernorm:
|
|
queries = self.q_layernorm(queries)
|
|
keys = self.k_layernorm(keys)
|
|
queries = queries.transpose(0, 2, 1, 3)
|
|
keys = keys.transpose(0, 2, 1, 3)
|
|
values = values.reshape(B, L, self.num_key_value_heads, -1).transpose(
|
|
0, 2, 1, 3
|
|
)
|
|
|
|
# Add RoPE to the queries and keys and combine them with the cache
|
|
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)
|
|
|
|
queries = queries.astype(mx.float32)
|
|
keys = keys.astype(mx.float32)
|
|
|
|
# Finally perform the attention computation
|
|
scale = math.sqrt(1 / queries.shape[-1])
|
|
output = mx.fast.scaled_dot_product_attention(
|
|
queries, keys, values, scale=scale, mask=mask
|
|
).astype(values.dtype)
|
|
output = output.transpose(0, 2, 1, 3).reshape(B, L, -1)
|
|
return self.o_proj(output)
|
|
|
|
|
|
class MLP(nn.Module):
|
|
def __init__(self, dim, hidden_dim):
|
|
super().__init__()
|
|
self.gate_proj = nn.Linear(dim, hidden_dim, bias=False)
|
|
self.down_proj = nn.Linear(hidden_dim, dim, bias=False)
|
|
self.up_proj = nn.Linear(dim, hidden_dim, bias=False)
|
|
|
|
def __call__(self, x) -> mx.array:
|
|
return self.down_proj(nn.silu(self.gate_proj(x)) * self.up_proj(x))
|
|
|
|
|
|
class DecoderLayer(nn.Module):
|
|
def __init__(self, config: ModelArgs):
|
|
super().__init__()
|
|
self.self_attn = Attention(config=config)
|
|
self.mlp = MLP(config.hidden_size, config.intermediate_size)
|
|
self.input_layernorm = nn.LayerNorm(
|
|
config.hidden_size,
|
|
eps=config.layer_norm_eps,
|
|
)
|
|
self.use_parallel_residual = config.use_parallel_residual
|
|
if not self.use_parallel_residual:
|
|
self.post_attention_layernorm = nn.LayerNorm(
|
|
config.hidden_size,
|
|
eps=config.layer_norm_eps,
|
|
)
|
|
|
|
def __call__(self, x, mask, cache):
|
|
h = self.input_layernorm(x)
|
|
r = self.self_attn(h, mask, cache)
|
|
|
|
if self.use_parallel_residual:
|
|
out = x + r + self.mlp(h)
|
|
else:
|
|
h = x + r
|
|
r = self.mlp(self.post_attention_layernorm(h))
|
|
out = h + r
|
|
return out
|
|
|
|
|
|
class StableLM(nn.Module):
|
|
def __init__(self, config: ModelArgs):
|
|
super().__init__()
|
|
self.embed_tokens = nn.Embedding(config.vocab_size, config.hidden_size)
|
|
self.layers = [DecoderLayer(config) for i in range(config.num_hidden_layers)]
|
|
self.norm = nn.LayerNorm(config.hidden_size, eps=config.layer_norm_eps)
|
|
|
|
def __call__(self, x, mask, cache):
|
|
x = self.embed_tokens(x)
|
|
if cache is None:
|
|
cache = [None] * len(self.layers)
|
|
|
|
for layer, c in zip(self.layers, cache):
|
|
x = layer(x, mask, cache=c)
|
|
|
|
return self.norm(x)
|
|
|
|
|
|
class Model(nn.Module):
|
|
def __init__(self, config: ModelArgs):
|
|
super().__init__()
|
|
self.model_type = config.model_type
|
|
self.model = StableLM(config)
|
|
self.lm_head = nn.Linear(config.hidden_size, config.vocab_size, bias=False)
|
|
self.args = config
|
|
|
|
def __call__(
|
|
self,
|
|
x: mx.array,
|
|
mask: mx.array = None,
|
|
cache: mx.array = None,
|
|
) -> Tuple[mx.array, mx.array]:
|
|
mask = create_attention_mask(x, cache)
|
|
y = self.model(x, mask, cache)
|
|
return self.lm_head(y)
|
|
|
|
@property
|
|
def layers(self):
|
|
return self.model.layers
|
|
|
|
@property
|
|
def head_dim(self):
|
|
return self.args.hidden_size // self.args.num_attention_heads
|
|
|
|
@property
|
|
def n_kv_heads(self):
|
|
return self.args.num_key_value_heads
|