File size: 6,644 Bytes
94ebc81
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
"""
StructuredReasoningDecoder: Cross-modal decoder that generates abstract reasoning outputs.
Cross-attends from graph node embeddings — never from raw text.
"""

import math
import torch
import torch.nn as nn
import torch.nn.functional as F
from typing import Optional


class GraphPositionalEncoding(nn.Module):
    def __init__(self, hidden_dim: int, max_nodes: int = 256):
        super().__init__()
        self.max_nodes = max_nodes
        self.node_pos_embed = nn.Embedding(max_nodes, hidden_dim)
        self.depth_embed = nn.Embedding(16, hidden_dim)
        nn.init.normal_(self.node_pos_embed.weight, std=0.02)
        nn.init.normal_(self.depth_embed.weight, std=0.02)

    def forward(self, graph_positions, tree_depths=None):
        pos_emb = self.node_pos_embed(graph_positions.clamp(0, self.max_nodes - 1))
        if tree_depths is not None:
            pos_emb = pos_emb + self.depth_embed(tree_depths.clamp(0, 15))
        return pos_emb


class CrossModalDecoderBlock(nn.Module):
    def __init__(self, config):
        super().__init__()
        self.hidden_dim = config.hidden_dim
        self.num_heads = config.num_heads
        self.head_dim = config.hidden_dim // config.num_heads
        self.scale = 1.0 / math.sqrt(self.head_dim)
        self.self_attn_qkv = nn.Linear(config.hidden_dim, config.hidden_dim * 3)
        self.self_attn_out = nn.Linear(config.hidden_dim, config.hidden_dim)
        self.cross_attn_q = nn.Linear(config.hidden_dim, config.hidden_dim)
        self.cross_attn_kv = nn.Linear(config.hidden_dim, config.hidden_dim * 2)
        self.cross_attn_out = nn.Linear(config.hidden_dim, config.hidden_dim)
        self.ffn = nn.Sequential(
            nn.Linear(config.hidden_dim, config.ffn_dim),
            nn.GELU(), nn.Dropout(config.dropout),
            nn.Linear(config.ffn_dim, config.hidden_dim),
        )
        self.ln1 = nn.LayerNorm(config.hidden_dim)
        self.ln2 = nn.LayerNorm(config.hidden_dim)
        self.ln3 = nn.LayerNorm(config.hidden_dim)
        self.dropout = nn.Dropout(config.dropout)

    def _self_attention(self, x, mask=None):
        B, L, D = x.shape
        qkv = self.self_attn_qkv(x).view(B, L, 3, self.num_heads, self.head_dim).permute(2, 0, 3, 1, 4)
        q, k, v = qkv[0], qkv[1], qkv[2]
        attn = torch.matmul(q, k.transpose(-2, -1)) * self.scale
        if mask is not None:
            attn = attn.masked_fill(mask == 0, float("-inf"))
        attn = F.softmax(attn, dim=-1)
        attn = self.dropout(attn)
        out = torch.matmul(attn, v).transpose(1, 2).contiguous().view(B, L, D)
        return self.self_attn_out(out)

    def _cross_attention(self, q_state, kv_state, mask=None):
        B, L, D = q_state.shape
        _, N, _ = kv_state.shape
        q = self.cross_attn_q(q_state).view(B, L, self.num_heads, self.head_dim).transpose(1, 2)
        kv = self.cross_attn_kv(kv_state).view(B, N, 2, self.num_heads, self.head_dim).permute(2, 0, 3, 1, 4)
        k, v = kv[0], kv[1]
        attn = torch.matmul(q, k.transpose(-2, -1)) * self.scale
        if mask is not None:
            attn = attn.masked_fill(mask.unsqueeze(1) == 0, float("-inf"))
        attn = F.softmax(attn, dim=-1)
        attn = self.dropout(attn)
        out = torch.matmul(attn, v).transpose(1, 2).contiguous().view(B, L, D)
        return self.cross_attn_out(out)

    def forward(self, x, encoder_hidden, self_mask=None, cross_mask=None):
        x = x + self.dropout(self._self_attention(self.ln1(x), self_mask))
        x = x + self.dropout(self._cross_attention(self.ln2(x), encoder_hidden, cross_mask))
        x = x + self.dropout(self.ffn(self.ln3(x)))
        return x


class StructuredReasoningDecoder(nn.Module):
    def __init__(self, config):
        super().__init__()
        self.config = config
        self.hidden_dim = config.hidden_dim
        self.vocab_size = config.vocab_size
        self.token_embed = nn.Embedding(config.vocab_size, config.hidden_dim)
        self.graph_pos_enc = GraphPositionalEncoding(config.hidden_dim, config.max_graph_nodes)
        self.layers = nn.ModuleList([
            CrossModalDecoderBlock(config) for _ in range(config.num_decoder_layers)
        ])
        self.final_ln = nn.LayerNorm(config.hidden_dim)
        self.lm_head = nn.Linear(config.hidden_dim, config.vocab_size, bias=False)
        self.dropout = nn.Dropout(config.dropout)
        nn.init.normal_(self.token_embed.weight, std=0.02)
        nn.init.normal_(self.lm_head.weight, std=0.02)

    def forward(self, input_ids, encoder_hidden, graph_positions=None,
                tree_depths=None, attention_mask=None, encoder_mask=None):
        B, L = input_ids.shape
        x = self.token_embed(input_ids)
        if graph_positions is not None:
            x = x + self.graph_pos_enc(graph_positions, tree_depths)
        causal_mask = torch.tril(torch.ones(L, L, device=input_ids.device)).view(1, 1, L, L)
        for layer in self.layers:
            x = layer(x, encoder_hidden, self_mask=causal_mask, cross_mask=encoder_mask)
        x = self.final_ln(x)
        return self.lm_head(x)

    @torch.no_grad()
    def generate(self, encoder_hidden, start_token_id, graph_positions=None,
                 encoder_mask=None, max_length=128, temperature=0.8,
                 top_k=50, top_p=0.9):
        B = encoder_hidden.shape[0]
        device = encoder_hidden.device
        generated = torch.full((B, 1), start_token_id, dtype=torch.long, device=device)
        for _ in range(max_length - 1):
            logits = self.forward(generated, encoder_hidden, None, None, None, encoder_mask)
            next_logits = logits[:, -1, :] / temperature
            if top_k > 0:
                v, _ = torch.topk(next_logits, top_k)
                next_logits[next_logits < v[:, -1:]] = float("-inf")
            if top_p < 1.0:
                sorted_logits, sorted_indices = torch.sort(next_logits, descending=True)
                cumulative_probs = torch.cumsum(F.softmax(sorted_logits, dim=-1), dim=-1)
                sorted_indices_to_remove = cumulative_probs > top_p
                sorted_indices_to_remove[..., 1:] = sorted_indices_to_remove[..., :-1].clone()
                sorted_indices_to_remove[..., 0] = False
                for b in range(B):
                    indices_to_remove = sorted_indices[b][sorted_indices_to_remove[b]]
                    next_logits[b, indices_to_remove] = float("-inf")
            probs = F.softmax(next_logits, dim=-1)
            next_token = torch.multinomial(probs, num_samples=1)
            generated = torch.cat([generated, next_token], dim=1)
        return generated