chore(mlx-lm): update phi2 model args to sync with hf config format. (#311)

* chore(mlx-lm): update phi2 model args to sync with hf config format

* chore: fix type hint

llms/README.md

@@ -61,7 +61,7 @@ text using the given prompt.
 For a full list of options run:
 
 ```
-python -m mlx_lm generate --help
+python -m mlx_lm.generate --help
 ```
 
 To quantize a model from the command line run:

llms/mlx_lm/models/phi2.py

@@ -10,12 +10,20 @@ from .base import BaseModelArgs
 
 @dataclass
 class ModelArgs(BaseModelArgs):
-    n_positions: int = 2048
+    max_position_embeddings: int = 2048
     vocab_size: int = 51200
-    n_embd: int = 2560
+    hidden_size: int = 2560
-    n_head: int = 32
+    num_attention_heads: int = 32
-    n_layer: int = 32
+    num_hidden_layers: int = 32
-    rotary_dim: int = 32
+    num_key_value_heads: int = 32
+    partial_rotary_factor: float = 0.4
+    intermediate_size: int = 10240
+    layer_norm_eps: float = 1e-5
+    rope_theta: float = 10000.0
+
+    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):
@@ -23,30 +31,66 @@ class LayerNorm(nn.LayerNorm):
         return super().__call__(x.astype(mx.float32)).astype(x.dtype)
 
 
-class RoPEAttention(nn.Module):
+class PhiAttention(nn.Module):
-    def __init__(self, dims: int, n_head: int, rotary_dim: int):
+    def __init__(self, config: ModelArgs):
         super().__init__()
 
-        self.n_head = n_head
+        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.rope_theta
+        self.partial_rotary_factor = config.partial_rotary_factor
 
-        self.q_proj = nn.Linear(dims, dims)
+        if (self.head_dim * self.num_heads) != self.hidden_size:
-        self.k_proj = nn.Linear(dims, dims)
+            raise ValueError(
-        self.v_proj = nn.Linear(dims, dims)
+                f"hidden_size must be divisible by num_heads (got `hidden_size`: {self.hidden_size}"
-        self.dense = nn.Linear(dims, dims)
+                f" and `num_heads`: {self.num_heads})."
+            )
 
-        self.rope = nn.RoPE(rotary_dim, traditional=False)
+        self.q_proj = nn.Linear(
+            self.hidden_size, self.num_heads * self.head_dim, bias=True
+        )
+        self.k_proj = nn.Linear(
+            self.hidden_size, self.num_key_value_heads * self.head_dim, bias=True
+        )
+        self.v_proj = nn.Linear(
+            self.hidden_size, self.num_key_value_heads * self.head_dim, bias=True
+        )
+        self.dense = nn.Linear(
+            self.num_heads * self.head_dim, 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, mask=None, cache=None):
         queries, keys, values = self.q_proj(x), self.k_proj(x), self.v_proj(x)
 
         # Extract some shapes
-        n_head = self.n_head
         B, L, D = queries.shape
 
         # Prepare the queries, keys and values for the attention computation
-        queries = queries.reshape(B, L, n_head, -1).transpose(0, 2, 1, 3)
+        queries = queries.reshape(B, L, self.num_heads, self.head_dim).transpose(
-        keys = keys.reshape(B, L, n_head, -1).transpose(0, 2, 1, 3)
+            0, 2, 1, 3
-        values = values.reshape(B, L, n_head, -1).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):
+            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:
@@ -74,25 +118,23 @@ class RoPEAttention(nn.Module):
         return self.dense(values_hat), (keys, values)
 
 
-class MLP(nn.Module):
+class PhiMLP(nn.Module):
-    def __init__(self, dim, hidden_dim):
+    def __init__(self, config: ModelArgs):
         super().__init__()
-        self.fc1 = nn.Linear(dim, hidden_dim)
+        self.fc1 = nn.Linear(config.hidden_size, config.intermediate_size)
-        self.fc2 = nn.Linear(hidden_dim, dim)
+        self.fc2 = nn.Linear(config.intermediate_size, config.hidden_size)
         self.act = nn.GELU(approx="precise")
 
     def __call__(self, x) -> mx.array:
         return self.fc2(self.act(self.fc1(x)))
 
 
-class ParallelBlock(nn.Module):
+class PhiDecoderLayer(nn.Module):
     def __init__(self, config: ModelArgs):
         super().__init__()
-        dims = config.n_embd
+        self.self_attn = PhiAttention(config=config)
-        mlp_dims = dims * 4
+        self.input_layernorm = LayerNorm(config.hidden_size, eps=config.layer_norm_eps)
-        self.self_attn = RoPEAttention(dims, config.n_head, config.rotary_dim)
+        self.mlp = PhiMLP(config)
-        self.input_layernorm = LayerNorm(dims)
-        self.mlp = MLP(dims, mlp_dims)
 
     def __call__(self, x, mask, cache):
         h = self.input_layernorm(x)
@@ -101,12 +143,12 @@ class ParallelBlock(nn.Module):
         return attn_h + ff_h + x, cache
 
 
-class Transformer(nn.Module):
+class PhiModel(nn.Module):
     def __init__(self, config: ModelArgs):
         super().__init__()
-        self.embed_tokens = nn.Embedding(config.vocab_size, config.n_embd)
+        self.embed_tokens = nn.Embedding(config.vocab_size, config.hidden_size)
-        self.layers = [ParallelBlock(config) for i in range(config.n_layer)]
+        self.layers = [PhiDecoderLayer(config) for i in range(config.num_hidden_layers)]
-        self.final_layernorm = LayerNorm(config.n_embd)
+        self.final_layernorm = LayerNorm(config.hidden_size, eps=config.layer_norm_eps)
 
     def __call__(self, x, mask, cache):
         x = self.embed_tokens(x)
@@ -121,8 +163,8 @@ class Transformer(nn.Module):
 class Model(nn.Module):
     def __init__(self, config: ModelArgs):
         super().__init__()
-        self.model = Transformer(config)
+        self.model = PhiModel(config)
-        self.lm_head = nn.Linear(config.n_embd, config.vocab_size)
+        self.lm_head = nn.Linear(config.hidden_size, config.vocab_size, bias=True)
 
     def __call__(
         self,