Add GPTNeoX model

llms/mlx_lm/models/gpt_neox.py

@@ -0,0 +1,202 @@
+import math
+from dataclasses import dataclass
+from typing import Tuple, Optional
+
+import mlx.core as mx
+import mlx.nn as nn
+
+from .base import BaseModelArgs
+
+
+@dataclass
+class ModelArgs(BaseModelArgs):
+    max_position_embeddings: int = 2048
+    vocab_size: int = 51200
+    hidden_size: int = 3072
+    num_attention_heads = 32
+    num_hidden_layers: int = 30
+    num_key_value_heads: Optional[int] = None
+    rotary_pct: float = 1.0  # partial_rotary_factor
+    intermediate_size: int = 12288
+    layer_norm_eps: float = 1e-5
+    rotary_emb_base: float = 10000.0  # rope_theta
+    hidden_dropout: float = 0.0
+
+    def __post_init__(self):
+        if self.num_key_value_heads is None:
+            self.num_key_value_heads = self.num_attention_heads
+
+
+class LayerNorm(nn.LayerNorm):
+    def __call__(self, x: mx.array) -> mx.array:
+        return super().__call__(x.astype(mx.float32)).astype(x.dtype)
+
+
+
+class GPTNeoXAttention(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.repeats = self.num_heads // self.num_key_value_heads
+        self.rope_theta = config.rotary_emb_base
+        self.partial_rotary_factor = config.rotary_pct
+
+        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.hidden_size, bias=True)
+        self.k_proj = nn.Linear(self.hidden_size, self.hidden_size, bias=True)
+        self.v_proj = nn.Linear(self.hidden_size, self.hidden_size, bias=True)
+
+        self.dense = nn.Linear(self.hidden_size, self.hidden_size, bias=True)
+        self.rope = nn.RoPE(
+            int(self.partial_rotary_factor * self.head_dim),
+            traditional=False,
+            base=self.rope_theta,
+        )
+
+    def __call__(
+        self,
+        x: mx.array,
+        mask: Optional[mx.array] = None,
+        cache: Optional[Tuple[mx.array]] = None,
+    ) -> Tuple[mx.array, Tuple[mx.array, mx.array]]:
+        queries = self.q_proj(x)
+        keys = self.k_proj(x)
+        values = self.v_proj(x)
+
+        # Extract some shapes
+        B, L, _ = queries.shape
+
+        # Prepare the queries, keys and values for the attention computation
+        queries = queries.reshape(B, L, self.num_heads, self.head_dim).transpose(0, 2, 1, 3)
+        keys = keys.reshape(B, L, self.num_key_value_heads, self.head_dim).transpose(0, 2, 1, 3)
+        values = values.reshape(B, L, self.num_key_value_heads, self.head_dim).transpose(0, 2, 1, 3)
+
+        def repeat(a: mx.array) -> mx.array:
+            a = mx.concatenate([mx.expand_dims(a, 2)] * self.repeats, axis=2)
+            return a.reshape([B, self.num_heads, L, -1])
+
+
+        if self.repeats > 1:
+            keys, values = map(repeat, (keys, values))
+
+        # Add RoPE to the queries and keys and combine them with the cache
+        if cache is not None:
+            key_cache, value_cache = cache
+            queries = self.rope(queries, offset=key_cache.shape[2])
+            keys = self.rope(keys, offset=key_cache.shape[2])
+            keys = mx.concatenate([key_cache, keys], axis=2)
+            values = mx.concatenate([value_cache, values], axis=2)
+        else:
+            queries = self.rope(queries)
+            keys = self.rope(keys)
+
+        queries = queries.astype(mx.float32)
+        keys = keys.astype(mx.float32)
+
+        # Finally preform the attention computation
+        scale = math.sqrt(1 / queries.shape[-1])
+        scores = (queries * scale) @ keys.transpose(0, 1, 3, 2)
+        if mask is not None:
+            scores = scores + mask
+
+        scores = mx.softmax(scores, axis=-1).astype(values.dtype)
+        values_hat = (scores @ values).transpose(0, 2, 1, 3).reshape(B, L, -1)
+
+        return self.dense(values_hat), (keys, values)
+
+
+class GPTNeoXMLP(nn.Module):
+    def __init__(self, config: ModelArgs):
+        super().__init__()
+        self.fc1 = nn.Linear(config.hidden_size, config.intermediate_size)
+        self.fc2 = nn.Linear(config.intermediate_size, config.hidden_size)
+        self.act = nn.GELU(approx="precise")
+
+    def __call__(self, x: mx.array) -> mx.array:
+        return self.fc2(self.act(self.fc1(x)))
+
+
+class GPTNeoXLayer(nn.Module):
+    def __init__(self, config: ModelArgs):
+        super().__init__()
+        self.input_layernorm = LayerNorm(
+            dims=config.hidden_size,
+            eps=config.layer_norm_eps,
+        )
+        self.attention = GPTNeoXAttention(config)
+        self.post_attention_layernorm = nn.LayerNorm(
+            dims=config.hidden_size,
+            eps=config.layer_norm_eps,
+        )
+        self.post_attention_dropout = nn.Dropout(config.hidden_dropout)
+        self.mlp = GPTNeoXMLP(config)
+        self.post_mlp_dropout = nn.Dropout(config.hidden_dropout)
+
+    def __call__(
+        self,
+        x: mx.array,
+        mask: Optional[mx.array] = None,
+        cache: Optional[Tuple[mx.array]] = None,
+    ) -> Tuple[mx.array, Tuple[mx.array, mx.array]]:
+        h = self.input_layernorm(x)
+        attn_layer_outputs = self.attention(h, mask, cache)
+        attn_output, cache = attn_layer_outputs[0], attn_layer_outputs[1]
+        attn_output = self.post_attention_dropout(attn_output)
+
+        attn_output = attn_output + x
+        mlp_output = self.mlp(self.post_attention_layernorm(attn_output))
+        mlp_output = self.post_mlp_dropout(mlp_output)
+        return mlp_output + attn_output, cache
+
+
+class GPTNeoXModel(nn.Module):
+    def __init__(self, config: ModelArgs):
+        super().__init__()
+        self.embed_tokens = nn.Embedding(config.vocab_size, config.hidden_size)
+        self.layers = [GPTNeoXLayer(config) for i in range(config.num_hidden_layers)]
+        self.final_layernorm = LayerNorm(dims=config.hidden_size, eps=config.layer_norm_eps)
+
+    def __call__(
+        self,
+        x: mx.array,
+        mask: Optional[mx.array] = None,
+        cache: Optional[Tuple[mx.array]] = None,
+    ) -> Tuple[mx.array, Tuple[mx.array]]:
+        x = self.embed_tokens(x)
+        if cache is None:
+            cache = [None] * len(self.layers)
+
+        for i, layers in enumerate(self.layers):
+            x, cache[i] = layers(x, mask, cache[i])
+
+        return self.final_layernorm(x), cache
+
+
+class Model(nn.Module):
+    def __init__(self, config: ModelArgs):
+        super().__init__()
+        self.model = GPTNeoXModel(config)
+        self.embed_out = nn.Linear(config.hidden_size, config.vocab_size, bias=True)
+
+    def __call__(
+        self,
+        x: mx.array,
+        mask: Optional[mx.array] = None,
+        cache: Optional[Tuple[mx.array]] = None,
+    ) -> Tuple[mx.array, Tuple[mx.array]]:
+        mask = None
+        if x.shape[1] > 1:
+            mask = nn.MultiHeadAttention.create_additive_causal_mask(x.shape[1])
+            mask = mask.astype(x.dtype)
+
+        y, cache = self.model(x, mask, cache)
+        return self.embed_out(y), cache

llms/mlx_lm/utils.py

@@ -10,7 +10,7 @@
 from transformers import AutoTokenizer, PreTrainedTokenizer
 
 # Local imports
-from .models import llama, mixtral, phi2
+from .models import llama, mixtral, phi2, gpt_neox
 from .models.base import BaseModelArgs
 
 # Constants
@@ -19,6 +19,7 @@
     "mistral": llama,  # mistral is compatible with llama
     "mixtral": mixtral,
     "phi": phi2,
+    "gpt_neox": gpt_neox,
 }
 
 linear_class_predicate = (