File size: 8,379 Bytes
34e468d
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
"""
A standalone, selectable GRU baseline, written to match the same interface as
the Transformer / Mamba models in this repo so it can be picked with `--model gru`.

Design notes (parity with model/mamba.py so the rest of the pipeline works
unchanged):
  * `self.layers` is a ModuleList of per-layer residual GRU blocks. The shared
    helper `get_block_list(model) = model.transformer.h if hasattr(model,
    'transformer') else model.layers` therefore returns an indexable list whose
    last element produces a (B, L, D) hidden state (so attention/activation
    hooks in the analysis scripts behave like they do for Mamba).
  * embedding weight is tied to lm_head (weight sharing), padding_idx=0.
  * `forward(idx, targets=None)` returns `(logits, loss)`; the loss uses
    `ignore_index=pad_id` and, at inference time (targets is None), only the last
    position is projected through lm_head.
  * `configure_optimizers`, `estimate_mfu`, `get_num_params` mirror Mamba.
"""

import math
import inspect
from dataclasses import dataclass
from typing import Optional

import torch
import torch.nn as nn
import torch.nn.functional as F


@dataclass
class GRUConfig:
    n_embd: int                      # D (hidden size of each GRU layer)
    n_layer: int
    vocab_size: int = 64
    dropout: float = 0.0
    bias: bool = True                # bias inside the GRU cells
    rms_norm_eps: float = 1e-5
    pad_id: int = -1
    model_type: str = "gru"


# taken straight from model/mamba.py (mamba-minimal) so normalization matches
class RMSNorm(nn.Module):
    def __init__(self, n_embd: int, eps: float = 1e-5):
        super().__init__()
        self.eps = eps
        self.weight = nn.Parameter(torch.ones(n_embd))

    def forward(self, x):
        return x * torch.rsqrt(x.pow(2).mean(-1, keepdim=True) + self.eps) * self.weight


class GRUBlock(nn.Module):
    """Pre-norm residual GRU layer: out = x + dropout(GRU(RMSNorm(x)))."""

    def __init__(self, config: GRUConfig):
        super().__init__()
        self.norm = RMSNorm(config.n_embd, config.rms_norm_eps)
        self.gru = nn.GRU(
            input_size=config.n_embd,
            hidden_size=config.n_embd,
            num_layers=1,
            batch_first=True,
            bias=config.bias,
        )
        self.dropout = nn.Dropout(config.dropout)

    def forward(self, x):
        # x : (B, L, D) -> (B, L, D)
        y, _ = self.gru(self.norm(x))
        return x + self.dropout(y)


class GRU(nn.Module):
    def __init__(self, config: GRUConfig):
        super().__init__()
        self.config = config

        self.embedding = nn.Embedding(config.vocab_size, config.n_embd, padding_idx=0)
        self.lm_head = nn.Linear(config.n_embd, config.vocab_size, bias=False)
        self.lm_head.weight = self.embedding.weight  # weight tying

        self.drop = nn.Dropout(config.dropout)
        self.layers = nn.ModuleList([GRUBlock(config) for _ in range(config.n_layer)])
        self.out_norm = RMSNorm(config.n_embd, config.rms_norm_eps)

        self.apply(self._init_weights)
        print(f"number of parameters: {self.get_num_params() / 1e6:.2f}M")

    def forward(self, idx, targets=None):
        # idx : (B, L)
        x = self.drop(self.embedding(idx))  # (B, L, D)
        for layer in self.layers:
            x = layer(x)
        x = self.out_norm(x)

        if targets is not None:
            logits = self.lm_head(x)
            loss = F.cross_entropy(
                logits.view(-1, logits.size(-1)),
                targets.view(-1),
                ignore_index=self.config.pad_id,
            )
        else:
            # inference-time mini-optimization: only project the last position
            logits = self.lm_head(x[:, [-1], :])
            loss = None

        return logits, loss

    @torch.no_grad()
    def generate(self, idx, max_new_tokens, temperature=1.0, top_k=None, return_confidence=False):
        """Autoregressively complete idx (B, T) by re-forwarding the full sequence each step.

        The GRU is recurrent and has no fixed context window, so no cropping is needed.
        Matches the return contract of model.transformer.GPT.generate:
            return_confidence=False -> idx
            return_confidence=True  -> (idx, confidences, top3_tokens, top3_probs)
        For B == 1 the confidence outputs are flat lists indexed by time step; for
        B > 1 they are per-sample lists of shape (B, T[, 3]).
        """
        confidences = [] if return_confidence else None
        top3_tokens = [] if return_confidence else None
        top3_probs = [] if return_confidence else None
        B = idx.size(0)

        for _ in range(max_new_tokens):
            logits, _ = self(idx)  # targets=None -> logits is (B, 1, V) for last position
            if temperature <= 0:
                # Greedy decoding (argmax); probs are the raw softmax for confidence reporting.
                probs = F.softmax(logits[:, -1, :], dim=-1)
                idx_next = probs.argmax(dim=-1, keepdim=True)  # (B, 1)
            else:
                logits = logits[:, -1, :] / temperature
                if top_k is not None:
                    v, _ = torch.topk(logits, min(top_k, logits.size(-1)))
                    logits[logits < v[:, [-1]]] = -float('Inf')
                probs = F.softmax(logits, dim=-1)
                idx_next = torch.multinomial(probs, num_samples=1)  # (B, 1)

            if return_confidence:
                sampled_probs = probs.gather(1, idx_next).squeeze(-1)  # (B,)
                confidences.append(sampled_probs.cpu().tolist())
                top3_prob_vals, top3_token_ids = torch.topk(probs, 3, dim=-1)  # (B, 3)
                top3_tokens.append(top3_token_ids.cpu().tolist())
                top3_probs.append(top3_prob_vals.cpu().tolist())

            idx = torch.cat((idx, idx_next), dim=1)

        if return_confidence:
            if B == 1:
                return (idx,
                        [c[0] for c in confidences],
                        [t[0] for t in top3_tokens],
                        [p[0] for p in top3_probs])
            T = len(confidences)
            conf_bs = [[confidences[t][b] for t in range(T)] for b in range(B)]
            tok_bs = [[top3_tokens[t][b] for t in range(T)] for b in range(B)]
            prob_bs = [[top3_probs[t][b] for t in range(T)] for b in range(B)]
            return idx, conf_bs, tok_bs, prob_bs
        return idx

    def _init_weights(self, module):
        if isinstance(module, nn.Linear):
            torch.nn.init.normal_(module.weight, mean=0.0, std=0.02)
            if module.bias is not None:
                torch.nn.init.zeros_(module.bias)
        elif isinstance(module, nn.Embedding):
            torch.nn.init.normal_(module.weight, mean=0.0, std=0.02)

    def configure_optimizers(self, weight_decay, learning_rate, betas, device_type):
        # all params that require grad; 2D+ tensors get weight decay, others don't.
        param_dict = {pn: p for pn, p in self.named_parameters() if p.requires_grad}
        decay_params = [p for p in param_dict.values() if p.dim() >= 2]
        nodecay_params = [p for p in param_dict.values() if p.dim() < 2]
        optim_groups = [
            {"params": decay_params, "weight_decay": weight_decay},
            {"params": nodecay_params, "weight_decay": 0.0},
        ]
        num_decay_params = sum(p.numel() for p in decay_params)
        num_nodecay_params = sum(p.numel() for p in nodecay_params)
        print(f"num decayed parameter tensors: {len(decay_params)}, with {num_decay_params:,} parameters")
        print(f"num non-decayed parameter tensors: {len(nodecay_params)}, with {num_nodecay_params:,} parameters")
        fused_available = "fused" in inspect.signature(torch.optim.AdamW).parameters
        use_fused = fused_available and device_type == "cuda"
        extra_args = dict(fused=True) if use_fused else {}
        optimizer = torch.optim.AdamW(optim_groups, lr=learning_rate, betas=betas, **extra_args)
        print(f"using fused AdamW: {use_fused}")
        return optimizer

    def estimate_mfu(self, fwdbwd_per_iter, dt):
        return -1

    def get_num_params(self, non_embedding=True):
        n_params = sum(p.numel() for p in self.parameters())
        if non_embedding:
            n_params -= self.embedding.weight.numel()
        return n_params