Move position biases to attention module

t5/convert.py

@@ -13,10 +13,6 @@ SHARED_REPLACEMENT_PATTERNS = [
     (".layer.1.layer_norm.", ".ln2."),
     (".layer.2.layer_norm.", ".ln3."),
     (".final_layer_norm.", ".ln."),
-    (
-        ".layers.0.layer.0.SelfAttention.relative_attention_bias.",
-        ".position_bias.relative_attention_bias."
-    ),
 ]
 
 ENCODER_REPLACEMENT_PATTERNS = [

t5/t5.py

@@ -1,6 +1,5 @@
 import argparse
 import math
-from typing import Optional
 from dataclasses import dataclass
 
 import mlx.core as mx
@@ -79,12 +78,9 @@ class RelativePositionBias(nn.Module):
         self.num_buckets = config.relative_attention_num_buckets
         self.max_distance = config.n_positions
         self.n_heads = config.num_heads
-        self.relative_attention_bias = nn.Embedding(self.num_buckets, self.n_heads)
 
-    def compute_bias(self, query_length, key_length, device=None):
+    def compute_bias(self, query_length, key_length):
         """Compute binned relative position bias"""
-        if device is None:
-            device = self.relative_attention_bias.weight.device
         context_position = mx.arange(query_length, dtype=mx.long)[:, None]
         memory_position = mx.arange(key_length, dtype=mx.long)[None, :]
         relative_position = memory_position - context_position  # shape (query_length, key_length)
@@ -100,35 +96,17 @@ class RelativePositionBias(nn.Module):
 
 
 class MultiHeadAttention(nn.Module):
-    def __init__(
-        self,
-        dims: int,
-        num_heads: int,
-        query_input_dims: Optional[int] = None,
-        key_input_dims: Optional[int] = None,
-        value_input_dims: Optional[int] = None,
-        value_dims: Optional[int] = None,
-        value_output_dims: Optional[int] = None,
-        bias: bool = False,
-    ):
+    def __init__(self, config: ModelArgs, has_relative_attention_bias: bool = False):
         super().__init__()
-
-        if (dims % num_heads) != 0:
-            raise ValueError(
-                f"The input feature dimensions should be divisible by the number of heads ({dims} % {num_heads}) != 0"
-            )
-
-        query_input_dims = query_input_dims or dims
-        key_input_dims = key_input_dims or dims
-        value_input_dims = value_input_dims or key_input_dims
-        value_dims = value_dims or dims
-        value_output_dims = value_output_dims or dims
-
-        self.num_heads = num_heads
-        self.query_proj = nn.Linear(query_input_dims, dims, bias=bias)
-        self.key_proj = nn.Linear(key_input_dims, dims, bias=bias)
-        self.value_proj = nn.Linear(value_input_dims, value_dims, bias=bias)
-        self.out_proj = nn.Linear(value_dims, value_output_dims, bias=bias)
+        self.num_heads = config.num_heads
+        self.query_proj = nn.Linear(config.d_model, config.d_model, bias=False)
+        self.key_proj = nn.Linear(config.d_model, config.d_model, bias=False)
+        self.value_proj = nn.Linear(config.d_model, config.d_model, bias=False)
+        self.out_proj = nn.Linear(config.d_model, config.d_model, bias=False)
+        if has_relative_attention_bias:
+            self.relative_attention_bias = nn.Embedding(
+                config.relative_attention_num_buckets,
+                config.num_heads)
 
     def __call__(self, queries, keys, values, mask=None):
         queries = self.query_proj(queries)
@@ -181,10 +159,12 @@ class LayerNorm(nn.Module):
 
 
 class TransformerEncoderLayer(nn.Module):
-    def __init__(self, config: ModelArgs):
+    def __init__(self, config: ModelArgs, has_relative_attention_bias: bool = False):
         super().__init__()
         mlp_dims = config.d_ff or config.d_model * 4
-        self.attention = MultiHeadAttention(config.d_model, config.num_heads)
+        self.attention = MultiHeadAttention(
+            config, has_relative_attention_bias=has_relative_attention_bias
+        )
         self.ln1 = LayerNorm(config.d_model, eps=config.layer_norm_epsilon)
         self.ln2 = LayerNorm(config.d_model, eps=config.layer_norm_epsilon)
         self.linear1 = nn.Linear(config.d_model, mlp_dims, bias=False)
@@ -208,11 +188,10 @@ class TransformerEncoder(nn.Module):
     def __init__(self, config: ModelArgs):
         super().__init__()
         self.layers = [
-            TransformerEncoderLayer(config)
-            for _ in range(config.num_layers)
+            TransformerEncoderLayer(config, has_relative_attention_bias=i == 0)
+            for i in range(config.num_layers)
         ]
         self.ln = LayerNorm(config.d_model, eps=config.layer_norm_epsilon)
-        self.position_bias = RelativePositionBias(config)
 
     def __call__(self, x, mask):
         for layer in self.layers:
@@ -223,11 +202,13 @@ class TransformerEncoder(nn.Module):
 
 
 class TransformerDecoderLayer(nn.Module):
-    def __init__(self, config: ModelArgs):
+    def __init__(self, config: ModelArgs, has_relative_attention_bias: bool = False):
         super().__init__()
         mlp_dims = config.d_ff or config.d_model * 4
-        self.self_attention = MultiHeadAttention(config.d_model, config.num_heads)
-        self.cross_attention = MultiHeadAttention(config.d_model, config.num_heads)
+        self.self_attention = MultiHeadAttention(
+            config, has_relative_attention_bias=has_relative_attention_bias
+        )
+        self.cross_attention = MultiHeadAttention(config)
         self.ln1 = LayerNorm(config.d_model, eps=config.layer_norm_epsilon)
         self.ln2 = LayerNorm(config.d_model, eps=config.layer_norm_epsilon)
         self.ln3 = LayerNorm(config.d_model, eps=config.layer_norm_epsilon)
@@ -256,11 +237,10 @@ class TransformerDecoder(nn.Module):
     def __init__(self, config: ModelArgs):
         super().__init__()
         self.layers = [
-            TransformerDecoderLayer(config)
-            for _ in range(config.num_layers)
+            TransformerDecoderLayer(config, has_relative_attention_bias=i == 0)
+            for i in range(config.num_layers)
         ]
         self.ln = LayerNorm(config.d_model, eps=config.layer_norm_epsilon)
-        self.position_bias = RelativePositionBias(config)
 
     def __call__(self, x, memory, x_mask, memory_mask):
         for layer in self.layers:
@@ -318,11 +298,18 @@ def load_model():
     weights = mx.load("weights.npz")
     current_weights = tree_flatten(model.parameters())
     weights_to_load = list(weights.items())
-    current_weights_keys = set(k for k, _ in current_weights)
-    weights_to_load_keys = set(k for k, _ in weights_to_load)
+    current_weights_dict = dict(current_weights)
+    current_weights_keys = set(current_weights_dict.keys())
+    weights_to_load_dict = dict(weights_to_load)
+    weights_to_load_keys = set(weights_to_load_dict.keys())
     print("Missing weights: ", sorted(current_weights_keys - weights_to_load_keys))
     print()
     print("Weights ignored: ", sorted(weights_to_load_keys - current_weights_keys))
+    for key in current_weights_keys & weights_to_load_keys:
+        if weights_to_load_dict[key].shape != current_weights_dict[key].shape:
+            print("Shape mismatch for key: ", key)
+            print("Expected shape: ", current_weights_dict[key].shape)
+            print("Loading shape: ", weights_to_load_dict[key].shape)
     model.update(tree_unflatten(weights_to_load))
     tokenizer = AutoTokenizer.from_pretrained("t5-small", trust_remote_code=True)
     return model, tokenizer