add phimoe

llms/mlx_lm/models/phimoe.py

@@ -0,0 +1,257 @@
+import math
+from dataclasses import dataclass
+from typing import List, Optional, Tuple, Union
+
+import mlx.core as mx
+import mlx.nn as nn
+
+from .base import BaseModelArgs, KVCache, create_attention_mask
+from .su_rope import SuScaledRotaryEmbedding
+
+
+@dataclass
+class ModelArgs(BaseModelArgs):
+    model_type: str = "phimoe"
+    vocab_size: int = 30000
+    hidden_size: int = 1024
+    intermediate_size: int = 4096
+    num_hidden_layers: int = 12
+    num_attention_heads: int = 16
+    num_key_value_heads: int = 16
+    max_position_embeddings: int = 2048
+    initializer_range: float = 0.02
+    rms_norm_eps: float = 1e-6
+    pad_token_id: Optional[int] = None
+    rope_traditional: bool = False
+    num_local_experts: int = 8
+    num_experts_per_tok: int = 2
+    attention_bias: bool = False
+    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
+
+        if self.rope_scaling:
+            required_keys = {"long_factor", "type"}
+            if not all(key in self.rope_scaling for key in required_keys):
+                raise ValueError(f"rope_scaling must contain keys {required_keys}")
+
+            if self.rope_scaling["type"] not in ["longrope", "su", "linear"]:
+                print(
+                    "[WARNING] rope_scaling 'type' currently only supports 'linear', 'su', and 'longrope'; setting rope scaling to false."
+                )
+                self.rope_scaling = None
+
+
+class Attention(nn.Module):
+    def __init__(self, args: ModelArgs):
+        super().__init__()
+
+        dim = args.hidden_size
+        self.n_heads = n_heads = args.num_attention_heads
+        assert args.num_key_value_heads is not None
+        self.n_kv_heads = n_kv_heads = args.num_key_value_heads
+
+        head_dim = args.hidden_size // n_heads
+        self.scale = head_dim**-0.5
+
+        self.q_proj = nn.Linear(dim, n_heads * head_dim, bias=True)
+        self.k_proj = nn.Linear(dim, n_kv_heads * head_dim, bias=True)
+        self.v_proj = nn.Linear(dim, n_kv_heads * head_dim, bias=True)
+        self.o_proj = nn.Linear(n_heads * head_dim, dim, bias=False)
+
+        rope_scale = 1.0
+        if args.rope_scaling and args.rope_scaling["type"] in ["longrope", "su"]:
+            self.rope = SuScaledRotaryEmbedding(
+                head_dim,
+                traditional=False,
+                base=args.rope_theta,
+                scale=rope_scale,
+                max_position_embeddings=args.max_position_embeddings,
+                original_max_position_embeddings=args.original_max_position_embeddings,
+                short_factor=args.rope_scaling["short_factor"],
+                long_factor=args.rope_scaling["long_factor"],
+            )
+        else:
+            if args.rope_scaling and args.rope_scaling["type"] == "linear":
+                assert isinstance(args.rope_scaling["factor"], float)
+                rope_scale = 1 / args.rope_scaling["factor"]
+            self.rope = nn.RoPE(
+                head_dim,
+                traditional=args.rope_traditional,
+                base=args.rope_theta,
+                scale=rope_scale,
+            )
+
+    def __call__(
+        self,
+        x: mx.array,
+        mask: Optional[mx.array] = None,
+        cache: Optional[KVCache] = None,
+    ) -> mx.array:
+        B, L, D = x.shape
+
+        queries, keys, values = self.q_proj(x), self.k_proj(x), self.v_proj(x)
+
+        # Prepare the queries, keys and values for the attention computation
+        queries = queries.reshape(B, L, self.n_heads, -1).transpose(0, 2, 1, 3)
+        keys = keys.reshape(B, L, self.n_kv_heads, -1).transpose(0, 2, 1, 3)
+        values = values.reshape(B, L, self.n_kv_heads, -1).transpose(0, 2, 1, 3)
+
+        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)
+
+        output = mx.fast.scaled_dot_product_attention(
+            queries, keys, values, scale=self.scale, mask=mask
+        )
+        output = output.transpose(0, 2, 1, 3).reshape(B, L, -1)
+        return self.o_proj(output)
+
+
+class PhiMoEBlockSparseTop2MLP(nn.Module):
+    def __init__(self, args: ModelArgs):
+        super().__init__()
+        self.ffn_dim = args.intermediate_size
+        self.hidden_dim = args.hidden_size
+
+        self.w1 = nn.Linear(self.hidden_dim, self.ffn_dim, bias=False)
+        self.w2 = nn.Linear(self.ffn_dim, self.hidden_dim, bias=False)
+        self.w3 = nn.Linear(self.hidden_dim, self.ffn_dim, bias=False)
+
+        self.act_fn = nn.GELU()
+
+    def __call__(self, hidden_states):
+        current_hidden_states = self.act_fn(self.w1(hidden_states)) * self.w3(
+            hidden_states
+        )
+        current_hidden_states = self.w2(current_hidden_states)
+        return current_hidden_states
+
+
+class PhiMoESparseMoeBlock(nn.Module):
+    def __init__(self, args: ModelArgs):
+        super().__init__()
+        self.hidden_dim = args.hidden_size
+        self.ffn_dim = args.intermediate_size
+        self.num_experts = args.num_local_experts
+        self.top_k = args.num_experts_per_tok
+
+        self.gate = nn.Linear(self.hidden_dim, self.num_experts, bias=False)
+        self.experts = [PhiMoEBlockSparseTop2MLP(args) for _ in range(self.num_experts)]
+
+    def __call__(self, hidden_states):
+        batch_size, sequence_length, hidden_dim = hidden_states.shape
+        hidden_states = hidden_states.reshape(-1, hidden_dim)
+
+        router_logits = self.gate(hidden_states)
+        routing_weights = mx.softmax(router_logits, axis=-1)
+        expert_indices = mx.argmax(routing_weights, axis=-1)
+
+        final_hidden_states = mx.zeros((batch_size * sequence_length, hidden_dim))
+
+        for expert_idx in range(self.num_experts):
+            expert_layer = self.experts[expert_idx]
+            expert_mask = expert_indices == expert_idx
+            if mx.sum(expert_mask) > 0:
+                expert_input = hidden_states[expert_mask]
+                expert_output = expert_layer(expert_input)
+                final_hidden_states = mx.where(
+                    expert_mask[:, None], expert_output, final_hidden_states
+                )
+
+        final_hidden_states = final_hidden_states.reshape(
+            batch_size, sequence_length, hidden_dim
+        )
+        return final_hidden_states, router_logits
+
+
+class PhiMoEDecoderLayer(nn.Module):
+    def __init__(self, args: ModelArgs):
+        super().__init__()
+        self.hidden_size = args.hidden_size
+
+        self.self_attn = Attention(args)
+        self.block_sparse_moe = PhiMoESparseMoeBlock(args)
+        self.input_layernorm = nn.LayerNorm(args.hidden_size, eps=args.rms_norm_eps)
+        self.post_attention_layernorm = nn.LayerNorm(
+            args.hidden_size, eps=args.rms_norm_eps
+        )
+
+    def __call__(self, hidden_states, attention_mask=None, position_ids=None):
+        residual = hidden_states
+        hidden_states = self.input_layernorm(hidden_states)
+
+        hidden_states = self.self_attn(
+            hidden_states=hidden_states,
+            attention_mask=attention_mask,
+            position_ids=position_ids,
+        )
+        hidden_states = residual + hidden_states
+
+        residual = hidden_states
+        hidden_states = self.post_attention_layernorm(hidden_states)
+        hidden_states, router_logits = self.block_sparse_moe(hidden_states)
+        hidden_states = residual + hidden_states
+
+        return hidden_states, router_logits
+
+
+class PhiMoEModel(nn.Module):
+    def __init__(self, args: ModelArgs):
+        super().__init__()
+        self.args = args
+        self.padding_idx = args.pad_token_id
+        self.vocab_size = args.vocab_size
+
+        self.embed_tokens = nn.Embedding(
+            args.vocab_size, args.hidden_size, self.padding_idx
+        )
+        self.layers = [PhiMoEDecoderLayer(args) for _ in range(args.num_hidden_layers)]
+        self.norm = nn.LayerNorm(args.hidden_size, eps=args.rms_norm_eps)
+
+    def __call__(self, input_ids, attention_mask=None, position_ids=None):
+        hidden_states = self.embed_tokens(input_ids)
+
+        for layer in self.layers:
+            hidden_states, _ = layer(hidden_states, attention_mask, position_ids)
+
+        hidden_states = self.norm(hidden_states)
+
+        return hidden_states
+
+
+class Model(nn.Module):
+    def __init__(self, args: ModelArgs):
+        super().__init__()
+        self.args = args
+        self.model = PhiMoEModel(args)
+        self.lm_head = nn.Linear(args.hidden_size, args.vocab_size, bias=False)
+
+    def __call__(self, input_ids, attention_mask=None, position_ids=None):
+        hidden_states = self.model(input_ids, attention_mask, position_ids)
+        logits = self.lm_head(hidden_states)
+        return logits
+
+    @property
+    def layers(self):
+        return self.model.layers
+
+    @property
+    def head_dim(self):
+        return self.args.hidden_size // self.args.num_attention_heads
+
+    def sanitize(self, weights):
+        # Remove unused precomputed rotary freqs
+        return {
+            k: v for k, v in weights.items() if "self_attn.rotary_emb.inv_freq" not in k
+        }
+
+    @property
+    def n_kv_heads(self):
+        return self.args.num_key_value_heads