File size: 2,207 Bytes
45b1630 370c6cd 45b1630 | 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 |
import torch, torch.nn as nn
from transformers import PreTrainedModel, PretrainedConfig
from transformers.modeling_outputs import CausalLMOutputWithPast
class ChessConfig(PretrainedConfig):
model_type = "chess_lm"
def __init__(self, vocab_size=1350, n_positions=256, n_embd=128, n_layer=4, n_head=4, n_ctx=256, tie_word_embeddings=True, **kwargs):
self.vocab_size, self.n_positions, self.n_embd = vocab_size, n_positions, n_embd
self.n_layer, self.n_head, self.n_ctx = n_layer, n_head, n_ctx
self.tie_word_embeddings = tie_word_embeddings
super().__init__(**kwargs)
class ChessForCausalLM(PreTrainedModel):
config_class = ChessConfig
def __init__(self, config):
super().__init__(config)
self.config = config
self.token_embedding = nn.Embedding(config.vocab_size, config.n_embd)
self.position_embedding = nn.Embedding(config.n_positions, config.n_embd)
encoder_layer = nn.TransformerEncoderLayer(d_model=config.n_embd, nhead=config.n_head, dim_feedforward=config.n_embd*4, batch_first=True, norm_first=True)
self.blocks = nn.TransformerEncoder(encoder_layer, num_layers=config.n_layer)
self.ln_f = nn.LayerNorm(config.n_embd)
self.head = nn.Linear(config.n_embd, config.vocab_size, bias=False)
if config.tie_word_embeddings: self.head.weight = self.token_embedding.weight
self.post_init()
def get_input_embeddings(self): return self.token_embedding
def set_input_embeddings(self, value): self.token_embedding = value
def forward(self, input_ids, labels=None, **kwargs):
B, T = input_ids.shape
x = self.token_embedding(input_ids) + self.position_embedding(torch.arange(T, device=input_ids.device))
mask = torch.triu(torch.ones(T, T, device=input_ids.device) * float('-inf'), diagonal=1)
x = self.ln_f(self.blocks(x, mask=mask, is_causal=True))
logits = self.head(x)
loss = None
if labels is not None:
loss = nn.CrossEntropyLoss()(logits[..., :-1, :].contiguous().view(-1, self.config.vocab_size), labels[..., 1:].contiguous().view(-1))
return CausalLMOutputWithPast(loss=loss, logits=logits)
|