nexaai 1.0.29__cp310-cp310-macosx_14_0_universal2.whl

nexaai/mlx_backend/vlm/modeling/models/llama4/language.py

@@ -0,0 +1,386 @@
+import inspect
+from dataclasses import dataclass
+from typing import Any, Optional
+
+import mlx.core as mx
+import mlx.nn as nn
+from mlx_lm.models.rope_utils import initialize_rope
+from mlx_lm.models.switch_layers import SwitchGLU
+
+from ..base import (
+    LanguageModelOutput,
+    create_attention_mask,
+    scaled_dot_product_attention,
+)
+from ..cache import ChunkedKVCache, KVCache
+
+
+@dataclass
+class TextConfig:
+    model_type: str
+    hidden_size: int
+    intermediate_size: int
+    num_attention_heads: int
+    rms_norm_eps: float
+    vocab_size: int
+    num_key_value_heads: int
+    rope_theta: float = 500000.0
+    num_hidden_layers: int = 48
+    rope_traditional: bool = False
+    rope_scaling: Optional[dict] = None  # Add missing rope_scaling attribute
+    tie_word_embeddings: bool = False
+    head_dim: int = 128
+    hidden_act: str = "silu"
+    intermediate_size_mlp: int = 16384
+    max_position_embeddings: int = 10485760
+    num_experts_per_tok: int = 1
+    num_local_experts: int = 16
+    attention_dropout: float = 0.0
+    use_qk_norm: bool = True
+    bos_token_id: int = 200000
+    eos_token_id: list = None
+    pad_token_id: int = 200018
+    attention_chunk_size: int = 8192
+    attention_bias: bool = False
+    interleave_moe_layer_step: int = 1
+    no_rope_layers: list = 4
+    output_router_logits: bool = False
+    router_aux_loss_coef: float = 0.001
+    router_jitter_noise: float = 0.0
+    attn_temperature_tuning: int = 4
+    floor_scale: float = 8192
+    attn_scale: float = 0.1
+    moe_layers: list = None
+
+    @classmethod
+    def from_dict(cls, params):
+        return cls(
+            **{
+                k: v
+                for k, v in params.items()
+                if k in inspect.signature(cls).parameters
+            }
+        )
+
+    def __post_init__(self):
+        if self.num_key_value_heads is None:
+            self.num_key_value_heads = self.num_attention_heads
+
+
+class Attention(nn.Module):
+    def __init__(self, config: TextConfig, layer_idx: int):
+        super().__init__()
+
+        dim = config.hidden_size
+        self.n_heads = n_heads = config.num_attention_heads
+        self.n_kv_heads = n_kv_heads = config.num_key_value_heads
+
+        self.use_rope = int((layer_idx + 1) % 4 != 0)  # rope unused for dense layers
+        self.attn_temperature_tuning = config.attn_temperature_tuning
+        self.floor_scale = config.floor_scale
+        self.attn_scale = config.attn_scale
+
+        self.head_dim = head_dim = config.head_dim or config.hidden_size // n_heads
+
+        self.scale = head_dim**-0.5
+        if hasattr(config, "attention_bias"):
+            attention_bias = config.attention_bias
+        else:
+            attention_bias = False
+
+        self.q_proj = nn.Linear(dim, n_heads * head_dim, bias=attention_bias)
+        self.k_proj = nn.Linear(dim, n_kv_heads * head_dim, bias=attention_bias)
+        self.v_proj = nn.Linear(dim, n_kv_heads * head_dim, bias=attention_bias)
+        self.o_proj = nn.Linear(n_heads * head_dim, dim, bias=attention_bias)
+
+        self.use_qk_norm = config.use_qk_norm and self.use_rope
+
+        if self.use_rope:
+            self.rope = initialize_rope(
+                head_dim,
+                config.rope_theta,
+                traditional=True,
+                scaling_config=config.rope_scaling,
+                max_position_embeddings=config.max_position_embeddings,
+            )
+
+    def __call__(
+        self,
+        x: mx.array,
+        mask: Optional[mx.array] = None,
+        cache: Optional[Any] = None,
+    ) -> mx.array:
+        B, L, D = x.shape
+
+        queries, keys, values = self.q_proj(x), self.k_proj(x), self.v_proj(x)
+
+        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:
+            offset = cache.offset
+        else:
+            offset = 0
+
+        if self.use_rope:
+            queries = self.rope(queries, offset=offset)
+            keys = self.rope(keys, offset=offset)
+
+        if self.use_qk_norm:
+            queries = mx.fast.rms_norm(queries, weight=None, eps=1e-6)
+            keys = mx.fast.rms_norm(keys, weight=None, eps=1e-6)
+
+        if self.attn_temperature_tuning and not self.use_rope:
+            attn_scales = (
+                mx.log(
+                    mx.floor(mx.arange(offset + 1, offset + L + 1) / self.floor_scale)
+                    + 1.0
+                )
+                * self.attn_scale
+                + 1.0
+            )
+            attn_scales = attn_scales[:, None]
+            queries = (queries * attn_scales).astype(queries.dtype)
+
+        if cache is not None:
+            keys, values = cache.update_and_fetch(keys, values)
+
+        if self.use_rope and isinstance(mask, mx.array):
+            key_len = keys.shape[-2]
+            if mask.shape[-1] != key_len:
+                mask = mask[..., -key_len:]
+
+        output = scaled_dot_product_attention(
+            queries, keys, values, cache, scale=self.scale, mask=mask
+        )
+        output = output.transpose(0, 2, 1, 3).reshape(B, L, -1)
+        return self.o_proj(output)
+
+
+class MLP(nn.Module):
+    def __init__(self, config: TextConfig, intermediate_size: int = None):
+        super().__init__()
+
+        dim = config.hidden_size
+        hidden_dim = intermediate_size or config.intermediate_size
+
+        self.gate_proj = nn.Linear(dim, hidden_dim, bias=False)
+        self.down_proj = nn.Linear(hidden_dim, dim, bias=False)
+        self.up_proj = nn.Linear(dim, hidden_dim, bias=False)
+
+    def __call__(self, x) -> mx.array:
+        return self.down_proj(nn.silu(self.gate_proj(x)) * self.up_proj(x))
+
+
+class MoE(nn.Module):
+    def __init__(self, config):
+        super().__init__()
+        self.top_k = config.num_experts_per_tok
+        self.num_experts = config.num_local_experts
+        self.experts = SwitchGLU(
+            config.hidden_size, config.intermediate_size, self.num_experts
+        )
+        self.router = nn.Linear(
+            config.hidden_size, config.num_local_experts, bias=False
+        )
+        self.shared_expert = MLP(config)
+
+    def __call__(self, x) -> mx.array:
+        logits = self.router(x)
+        k = self.top_k
+        indices = mx.argpartition(-logits, kth=k - 1, axis=-1)[..., :k]
+        scores = mx.take_along_axis(logits, indices, axis=-1)
+        scores = mx.sigmoid(scores.astype(mx.float32)).astype(x.dtype)
+
+        out = self.experts(x * scores, indices).squeeze(2)
+        return out + self.shared_expert(x)
+
+
+class TransformerBlock(nn.Module):
+    def __init__(self, config: TextConfig, layer_idx: int):
+        super().__init__()
+        self.num_attention_heads = config.num_attention_heads
+        self.hidden_size = config.hidden_size
+        self.self_attn = Attention(config, layer_idx)
+        self.use_chunked_attention = int((layer_idx + 1) % 4 != 0)
+        self.is_moe_layer = (layer_idx % config.interleave_moe_layer_step) == (
+            config.interleave_moe_layer_step - 1
+        )
+        if self.is_moe_layer:
+            self.feed_forward = MoE(config)
+        else:
+            self.feed_forward = MLP(config, config.intermediate_size_mlp)
+
+        self.input_layernorm = nn.RMSNorm(config.hidden_size, eps=config.rms_norm_eps)
+        self.post_attention_layernorm = nn.RMSNorm(
+            config.hidden_size, eps=config.rms_norm_eps
+        )
+        self.config = config
+
+        self.use_chunked_attention = int((layer_idx + 1) % 4 != 0)  # <=> use rope
+
+    def __call__(
+        self,
+        x: mx.array,
+        mask: Optional[mx.array] = None,
+        cache: Optional[Any] = None,
+    ) -> mx.array:
+
+        r = self.self_attn(self.input_layernorm(x), mask, cache)
+        h = x + r
+        r = self.feed_forward(self.post_attention_layernorm(h))
+        out = h + r
+        return out
+
+
+class LlamaModel(nn.Module):
+    def __init__(self, config: TextConfig):
+        super().__init__()
+        self.config = config
+        self.vocab_size = config.vocab_size
+        self.num_hidden_layers = config.num_hidden_layers
+        assert self.vocab_size > 0
+        self.embed_tokens = nn.Embedding(config.vocab_size, config.hidden_size)
+        self.layers = [
+            TransformerBlock(config, i) for i in range(config.num_hidden_layers)
+        ]
+        self.norm = nn.RMSNorm(config.hidden_size, eps=config.rms_norm_eps)
+
+    def create_chunked_attention_mask(
+        self, seq_len: int, attention_chunk_size: int, start: int = 0, offset: int = 0
+    ) -> mx.array:
+        """
+        Generate the following:
+
+        'What'      :  0 ■ ⬚ ⬚ ⬚ ⬚ ⬚    |
+        '▁is'       :  1 ■ ■ ⬚ ⬚ ⬚ ⬚     |
+        '▁ch'       :  2 ■ ■ ■ ⬚ ⬚ ⬚     |
+        'unked'     :  3 ⬚ ⬚ ⬚ ■ ⬚ ⬚    |
+        '▁attention':  4 ⬚ ⬚ ⬚ ■ ■ ⬚    |
+        '?'         :  5 ⬚ ⬚ ⬚ ■ ■ ■     |
+
+        If the chunk size is 3.
+        This can just be appplied over the already created attention mask
+        """
+
+        end = offset + seq_len
+        linds = mx.arange(start, end)
+        rinds = mx.arange(offset, end)[:, None]
+        block_pos = mx.abs(
+            (linds // attention_chunk_size) - (rinds // attention_chunk_size)
+        )
+        token_pos = linds <= rinds
+        mask = (block_pos == 0) & (token_pos)
+        return mask
+
+    def __call__(
+        self,
+        input_ids: mx.array = None,
+        input_embeds: mx.array = None,
+        mask: mx.array = None,
+        cache=None,
+    ):
+        if input_embeds is None:
+            h = self.embed_tokens(input_ids)
+        else:
+            h = input_embeds
+
+        if mask is None:
+            mask = create_attention_mask(h, cache)
+
+        if cache is not None:
+            for idx, c in enumerate(cache):
+                if (idx + 1) % 4 != 0:
+                    c.maybe_trim_front()
+            start = cache[0].start_position
+            offset = cache[0].offset
+        else:
+            start = 0
+            offset = 0
+
+        # Create a mask for the chunked attention
+        chunk_mask = self.create_chunked_attention_mask(
+            h.shape[1], self.config.attention_chunk_size, start, offset
+        )
+
+        if cache is None:
+            cache = [None] * len(self.layers)
+
+        for idx, (layer, c) in enumerate(zip(self.layers, cache)):
+            use_chunked_attention = (idx + 1) % 4 != 0
+            if use_chunked_attention:
+                local_mask = chunk_mask
+            else:
+                local_mask = mask
+            h = layer(h, local_mask, cache=c)
+
+        return self.norm(h)
+
+
+class LanguageModel(nn.Module):
+    def __init__(self, config: TextConfig):
+        super().__init__()
+        self.config = config
+        self.model_type = config.model_type
+        self.model = LlamaModel(self.config)
+        self.lm_head = nn.Linear(
+            self.config.hidden_size, self.config.vocab_size, bias=False
+        )
+
+    def __call__(
+        self,
+        input_ids: mx.array = None,
+        input_embeds: mx.array = None,
+        mask: mx.array = None,
+        cache=None,
+    ):
+        out = self.model(
+            input_ids=input_ids,
+            input_embeds=input_embeds,
+            mask=mask,
+            cache=cache,
+        )
+        out = self.lm_head(out)
+        return LanguageModelOutput(logits=out)
+
+    def sanitize(self, weights):
+        # Rename expert weights for SwitchGLU
+        for l in range(self.config.num_hidden_layers):
+            prefix = f"language_model.model.layers.{l}.feed_forward.experts"
+            if f"{prefix}.gate_up_proj" in weights:
+                v = weights.pop(f"{prefix}.gate_up_proj")
+                gate_k = f"{prefix}.gate_proj.weight"
+                up_k = f"{prefix}.up_proj.weight"
+                gate_proj, up_proj = mx.split(v, 2, axis=-1)
+                weights[gate_k] = mx.swapaxes(gate_proj, 1, 2)
+                weights[up_k] = mx.swapaxes(up_proj, 1, 2)
+            if f"{prefix}.down_proj" in weights:
+                down_proj = weights.pop(f"{prefix}.down_proj")
+                weights[f"{prefix}.down_proj.weight"] = mx.swapaxes(down_proj, 1, 2)
+        return weights
+
+    @property
+    def layers(self):
+        return self.model.layers
+
+    @property
+    def n_kv_heads(self):
+        return self.config.num_key_value_heads
+
+    @property
+    def head_dim(self):
+        return (
+            self.config.head_dim
+            if self.config.head_dim
+            else self.config.hidden_size // self.config.num_attention_heads
+        )
+
+    def make_cache(self):
+        caches = []
+        for i in range(self.config.num_hidden_layers):
+            if (i + 1) % 4 != 0:
+                caches.append(ChunkedKVCache(self.config.attention_chunk_size))
+            else:
+                caches.append(KVCache())  # no chunking for dense layers
+        return caches

nexaai/mlx_backend/vlm/modeling/models/llama4/llama4.py

@@ -0,0 +1,138 @@
+import glob
+import inspect
+import json
+import re
+from dataclasses import dataclass
+from pathlib import Path
+from typing import Any, Callable, List, Optional, Tuple, Union
+
+import mlx.core as mx
+import mlx.nn as nn
+import numpy as np
+
+from .language import LanguageModel, TextConfig
+from .vision import Llama4MultiModalProjector, VisionConfig, VisionModel
+
+
+@dataclass
+class ModelConfig:
+    text_config: TextConfig
+    vision_config: VisionConfig
+    model_type: str
+    ignore_index: int = -100
+    image_token_id: int = 200092
+    image_token_index: Optional[int] = None
+    eos_token_id: Optional[List[int]] = None
+
+    def __post_init__(self):
+        if self.image_token_index is None:
+            self.image_token_index = self.image_token_id
+
+    @classmethod
+    def from_dict(cls, params):
+        return cls(
+            **{
+                k: v
+                for k, v in params.items()
+                if k in inspect.signature(cls).parameters
+            }
+        )
+
+
+class Model(nn.Module):
+    def __init__(self, config: ModelConfig):
+        super().__init__()
+        self.config = config
+        self.vision_model = VisionModel(config.vision_config)
+        self.multi_modal_projector = Llama4MultiModalProjector(config)
+        self.language_model = LanguageModel(config.text_config)
+        self.vocab_size = config.text_config.vocab_size
+
+    def set_input_embeddings(self, value):
+        self.language_model.set_input_embeddings(value)
+
+    def get_output_embeddings(self):
+        return self.language_model.get_output_embeddings()
+
+    def set_output_embeddings(self, new_embeddings):
+        self.language_model.set_output_embeddings(new_embeddings)
+
+    def set_decoder(self, decoder):
+        self.language_model.set_decoder(decoder)
+
+    def get_decoder(self):
+        return self.language_model.get_decoder()
+
+    def get_image_features(
+        self,
+        pixel_values: mx.array,
+        vision_feature_layer: Union[int, List[int]],
+        vision_feature_select_strategy: str,
+        **kwargs,
+    ):
+        if vision_feature_select_strategy not in ["default", "full"]:
+            raise ValueError(
+                f"Unexpected select feature strategy: {self.vision_feature_select_strategy}"
+            )
+        kwargs = {k: v for k, v in kwargs.items() if v is not None}
+        hidden_state = self.vision_model(
+            pixel_values, output_hidden_states=False, **kwargs
+        )
+        return hidden_state
+
+    def get_input_embeddings(
+        self,
+        input_ids: Optional[mx.array] = None,
+        pixel_values: Optional[mx.array] = None,
+        **kwargs,
+    ):
+        if pixel_values is None:
+            return self.language_model.model.embed_tokens(input_ids)
+
+        # Get the input embeddings from the language model
+        inputs_embeds = self.language_model.model.embed_tokens(input_ids)
+
+        image_features = self.get_image_features(
+            pixel_values=pixel_values,
+            vision_feature_layer=kwargs.get("vision_feature_layer", -1),
+            vision_feature_select_strategy=kwargs.get(
+                "vision_feature_select_strategy", "default"
+            ),
+        )
+
+        vision_flat = image_features.reshape(-1, image_features.shape[-1])
+        projected_vision_flat = self.multi_modal_projector(vision_flat)
+
+        # Insert special image tokens in the input_ids
+        final_inputs_embeds = self._prepare_inputs_for_multimodal(
+            projected_vision_flat, inputs_embeds, input_ids
+        )
+        return final_inputs_embeds
+
+    def _prepare_inputs_for_multimodal(self, image_features, inputs_embeds, input_ids):
+        image_token_index = self.config.image_token_index
+
+        # Positions of <image> tokens in input_ids, assuming batch size is 1
+        image_positions = np.where(input_ids == image_token_index)[1].tolist()
+
+        inputs_embeds[:, image_positions, :] = image_features
+
+        return inputs_embeds
+
+    @property
+    def layers(self):
+        return self.language_model.model.layers
+
+    def __call__(
+        self,
+        input_ids: mx.array,
+        pixel_values: mx.array,
+        cache=None,
+        **kwargs,
+    ):
+
+        input_embeddings = self.get_input_embeddings(input_ids, pixel_values, **kwargs)
+        logits = self.language_model(
+            input_ids=input_ids, cache=cache, input_embeds=input_embeddings
+        )
+        return logits