Llama: support for `max_position_embeddings` (#22471)

* Llama now supports max_position_embeddings

* Save config; Cosmetic edits

src/transformers/models/llama/configuration_llama.py

@@ -52,6 +52,9 @@ class LlamaConfig(PretrainedConfig):
             Number of attention heads for each attention layer in the Transformer encoder.
         hidden_act (`str` or `function`, *optional*, defaults to `"silu"`):
             The non-linear activation function (function or string) in the decoder.
+        max_position_embeddings (`int`, *optional*, defaults to 2048):
+            The maximum sequence length that this model might ever be used with. Typically set this to something large
+            just in case (e.g., 512 or 1024 or 2048).
         initializer_range (`float`, *optional*, defaults to 0.02):
             The standard deviation of the truncated_normal_initializer for initializing all weight matrices.
         rms_norm_eps (`float`, *optional*, defaults to 1e-12):
@@ -85,6 +88,7 @@ class LlamaConfig(PretrainedConfig):
         num_hidden_layers=32,
         num_attention_heads=32,
         hidden_act="silu",
+        max_position_embeddings=2048,
         initializer_range=0.02,
         rms_norm_eps=1e-6,
         use_cache=True,
@@ -95,6 +99,7 @@ class LlamaConfig(PretrainedConfig):
         **kwargs,
     ):
         self.vocab_size = vocab_size
+        self.max_position_embeddings = max_position_embeddings
         self.hidden_size = hidden_size
         self.intermediate_size = intermediate_size
         self.num_hidden_layers = num_hidden_layers

src/transformers/models/llama/modeling_llama.py

@@ -160,42 +160,24 @@ class LlamaMLP(nn.Module):
 class LlamaAttention(nn.Module):
     """Multi-headed attention from 'Attention Is All You Need' paper"""
 
-    def __init__(
-        self,
-        hidden_size: int,
-        num_heads: int,
-    ):
+    def __init__(self, config: LlamaConfig):
         super().__init__()
-        self.hidden_size = hidden_size
-        self.num_heads = num_heads
-        self.head_dim = hidden_size // num_heads
+        self.config = config
+        self.hidden_size = config.hidden_size
+        self.num_heads = config.num_attention_heads
+        self.head_dim = self.hidden_size // self.num_heads
+        self.max_position_embeddings = config.max_position_embeddings
 
-        if (self.head_dim * num_heads) != self.hidden_size:
+        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`: {num_heads})."
+                f" and `num_heads`: {self.num_heads})."
             )
-        self.q_proj = nn.Linear(
-            hidden_size,
-            num_heads * self.head_dim,
-            bias=False,
-        )
-        self.k_proj = nn.Linear(
-            hidden_size,
-            num_heads * self.head_dim,
-            bias=False,
-        )
-        self.v_proj = nn.Linear(
-            hidden_size,
-            num_heads * self.head_dim,
-            bias=False,
-        )
-        self.o_proj = nn.Linear(
-            num_heads * self.head_dim,
-            hidden_size,
-            bias=False,
-        )
-        self.rotary_emb = LlamaRotaryEmbedding(self.head_dim)
+        self.q_proj = nn.Linear(self.hidden_size, self.num_heads * self.head_dim, bias=False)
+        self.k_proj = nn.Linear(self.hidden_size, self.num_heads * self.head_dim, bias=False)
+        self.v_proj = nn.Linear(self.hidden_size, self.num_heads * self.head_dim, bias=False)
+        self.o_proj = nn.Linear(self.num_heads * self.head_dim, self.hidden_size, bias=False)
+        self.rotary_emb = LlamaRotaryEmbedding(self.head_dim, max_position_embeddings=self.max_position_embeddings)
 
     def _shape(self, tensor: torch.Tensor, seq_len: int, bsz: int):
         return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
@@ -270,10 +252,7 @@ class LlamaDecoderLayer(nn.Module):
     def __init__(self, config: LlamaConfig):
         super().__init__()
         self.hidden_size = config.hidden_size
-        self.self_attn = LlamaAttention(
-            hidden_size=self.hidden_size,
-            num_heads=config.num_attention_heads,
-        )
+        self.self_attn = LlamaAttention(config=config)
         self.mlp = LlamaMLP(
             hidden_size=self.hidden_size,
             intermediate_size=config.intermediate_size,