from transformers import AutoConfig, PreTrainedModel, AutoModelForCausalLM from typing import List, Optional from torch import nn from model import LayerNorm, BlockJ from transformers.modeling_outputs import CausalLMOutputWithPast import torch import math from torch.nn import functional as F from transformers import AutoConfig, AutoModel from .pretrained_config import * class GPTJXForCausalLM(PreTrainedModel): config_class = GPTJXConfig base_model_prefix = "transformer" is_parallelizable = True supports_gradient_checkpointing = True _no_split_modules = ["BlockJ"] # _skip_keys_device_placement = "past_key_values" _supports_flash_attn_2 = True _tied_weights_keys = ["lm_head.weight"] def __init__(self, config): super().__init__(config) assert config.vocab_size is not None assert config.block_size is not None self.config = config self.transformer = nn.ModuleDict(dict( wte = nn.Embedding(config.vocab_size, config.n_embd), wpe = nn.Embedding(config.block_size, config.n_embd), drop = nn.Dropout(config.dropout), h = nn.ModuleList([BlockJ(config) for _ in range(config.n_layer)]), ln_f = LayerNorm(config.n_embd, bias=config.bias), )) self.lm_head = nn.Linear(config.n_embd, config.vocab_size, bias=False) # with weight tying when using torch.compile() some warnings get generated: # "UserWarning: functional_call was passed multiple values for tied weights. # This behavior is deprecated and will be an error in future versions" # not 100% sure what this is, so far seems to be harmless. TODO investigate self.transformer.wte.weight = self.lm_head.weight # https://paperswithcode.com/method/weight-tying # init all weights self.apply(self._init_weights) # apply special scaled init to the residual projections, per GPT-2 paper for pn, p in self.named_parameters(): if pn.endswith('c_proj.weight'): torch.nn.init.normal_(p, mean=0.0, std=0.02/math.sqrt(2 * config.n_layer)) # report number of parameters print("number of parameters: %.2fM" % (self.get_num_params()/1e6,)) def get_num_params(self, non_embedding=True): """ Return the number of parameters in the model. For non-embedding count (default), the position embeddings get subtracted. The token embeddings would too, except due to the parameter sharing these params are actually used as weights in the final layer, so we include them. """ n_params = sum(p.numel() for p in self.parameters()) if non_embedding: n_params -= self.transformer.wpe.weight.numel() return n_params def get_input_embeddings(self): return self.wte def set_input_embeddings(self, new_embeddings): self.wte = new_embeddings def forward(self, idx, targets=None, attn_mask= None, output_hidden_states: Optional[bool] = None, **kwargs): device = idx.device b, t = idx.size() # attn_mask = _prepare_mask_(idx, b, eval) # print("attention mask: ", attn_mask) assert t <= self.config.block_size, f"Cannot forward sequence of length {t}, block size is only {self.config.block_size}" pos = torch.arange(0, t, dtype=torch.long, device=device) # shape (t) # forward the GPT model itself tok_emb = self.transformer.wte(idx) # token embeddings of shape (b, t, n_embd) pos_emb = self.transformer.wpe(pos) # position embeddings of shape (t, n_embd) x = self.transformer.drop(tok_emb + pos_emb) for block in self.transformer.h: x = block(x, attn_mask=attn_mask) x = self.transformer.ln_f(x) if targets is not None: # if we are given some desired targets also calculate the loss logits = self.lm_head(x) loss = F.cross_entropy(logits.view(-1, logits.size(-1)), targets.view(-1), ignore_index=-100) else: # inference-time mini-optimization: only forward the lm_head on the very last position logits = self.lm_head(x[:, [-1], :]) # note: using list [-1] to preserve the time dim loss = None # return {"logits": logits, "loss": loss} return CausalLMOutputWithPast( loss=loss, logits=logits, hidden_states=x if output_hidden_states else None, attentions= None, ) def prepare_inputs_for_generation(self, input_ids, attention_mask=None, **kwargs): # Default model inputs model_inputs = {"idx": input_ids} # Add attention mask if provided if attention_mask is not None: model_inputs["attn_mask"] = attention_mask return model_inputs @torch.no_grad() def stream(self, idx, max_new_tokens, temperature=1.0, top_k=None,gen_mode="greedy"): """ Take a conditioning sequence of indices idx (LongTensor of shape (b,t)) and complete the sequence max_new_tokens times, feeding the predictions back into the model each time. Most likely you'll want to make sure to be in model.eval() mode of operation for this. """ for _ in range(max_new_tokens): # if the sequence context is growing too long we must crop it at block_size idx_cond = idx if idx.size(1) <= self.config.block_size else idx[:, -self.config.block_size:] # forward the model to get the logits for the index in the sequence logits, _ = self(idx_cond, eval=True) # pluck the logits at the final step and scale by desired temperature logits = logits[:, -1, :] / temperature # optionally crop the logits to only the top k options if top_k is not None: v, _ = torch.topk(logits, min(top_k, logits.size(-1))) logits[logits < v[:, [-1]]] = -float('Inf') # apply softmax to convert logits to (normalized) probabilities probs = F.softmax(logits, dim=-1) # sample from the distribution if gen_mode == 'greedy': idx_next = torch.argmax(probs, dim=-1).unsqueeze(0) else: idx_next = torch.multinomial(probs, num_samples=1) # print(idx_next.shape, idx.shape) idx = torch.cat((idx, idx_next), dim=1) # append sampled index to the running sequence and continue yield idx_next def crop_block_size(self, block_size): # model surgery to decrease the block size if necessary # e.g. we may load the GPT2 pretrained model checkpoint (block size 1024) # but want to use a smaller block size for some smaller, simpler model assert block_size <= self.config.block_size self.config.block_size = block_size self.transformer.wpe.weight = nn.Parameter(self.transformer.wpe.weight[:block_size]) for block in self.transformer.h: if hasattr(block.attn, 'bias'): block.attn.bias = block.attn.bias[:,:,:block_size,:block_size] AutoConfig.register("nanogpt-j", GPTJXConfig) AutoModel.register(GPTJXConfig,GPTJXForCausalLM) AutoModelForCausalLM.register(GPTJXConfig, GPTJXForCausalLM) if __name__ == '__main__': from transformers import AutoTokenizer tokenizer = AutoTokenizer.from_pretrained("BeardedMonster/SabiYarn") input_ids = tokenizer("ba wo ni?", return_tensors="pt")["input_ids"] targets = input_ids # config = GPTJConfig() # config.save_pretrained("gptj-config") # new_config = GPTJ.from_pretrained("gptj-config") # model = GPTJ(config) # state_dict = torch.load('model.pt', map_location="cpu") # model.load_state_dict(state_dict) model = GPTJXForCausalLM.from_pretrained("/pretrainedmodel") # model.save_pretrained("/pretrainedmodel") outputs = model(input_ids, targets) print(outputs) output = model.generate(input_ids, max_new_tokens=100) print(tokenizer.decode(output[0])) # print(new_config)