import os gpt_neo_series_id = '1.3B_ckpt' os.environ['gpt_neo_series_id'] = gpt_neo_series_id import torch import torch.nn as nn from methods.elasticdnn.api.algs.fm_lora import ElasticDNN_FMLoRAAlg from methods.elasticdnn.api.algs.md_pretraining_wo_fbs import ElasticDNN_MDPretrainingWoFBSAlg from methods.elasticdnn.model.base import ElasticDNNUtil from methods.elasticdnn.pipeline.offline.fm_lora.base import FMLoRA_Util from gpt_neo import getTokenizer, ElasticGPTUtil, FMLoRA_GPT_Util, ElasticDNN_OfflineTextGenFMModel, ElasticDNN_OfflineTextGenMDModel, FM_to_MD_GPT_Util, collate_fn from methods.elasticdnn.pipeline.offline.fm_to_md.base import FM_to_MD_Util from methods.elasticdnn.pipeline.offline.fm_to_md.vit import FM_to_MD_ViT_Util from methods.elasticdnn.model.vit import ElasticViTUtil from methods.elasticdnn.api.algs.md_pretraining_index_v2_train_index_and_md import ElasticDNN_MDPretrainingIndexAlg from utils.dl.common.model import LayerActivation2, get_module, get_parameter from utils.common.exp import save_models_dict_for_init, get_res_save_dir from data import build_gen_scenario import torch.nn.functional as F import os from utils.dl.common.loss import CrossEntropyLossSoft from new_impl.cv.feat_align.main_gpt_neo import OnlineFeatAlignModel, FeatAlignAlg import tqdm from new_impl.cv.feat_align.mmd import mmd_rbf from new_impl.cv.utils.elasticfm_da import init_online_model, elasticfm_da from methods.elasticdnn.api.online_model_v2 import ElasticDNN_OnlineModel from utils.common.log import logger import nltk from nltk.translate.bleu_score import sentence_bleu, corpus_bleu from nltk.translate.bleu_score import SmoothingFunction import json os.environ['TOKENIZERS_PARALLELISM'] = 'true' device = 'cuda:1' app_name = 'cls' sd_sparsity = 0.8 settings = { 'involve_fm': True } torch.cuda.set_device(1) scenario = build_gen_scenario( source_datasets_name=['No_robots'], target_datasets_order=['Medicine_task', 'Law_task'] * 10, da_mode='close_set', data_dirs={ 'No_robots': '/data/zql/datasets/no_robots', 'Law_task': '/data/zql/datasets/law_task', 'Medicine_task': '/data/zql/datasets/medicine_task', }, ) class ElasticDNN_TxtgenOnlineModel(ElasticDNN_OnlineModel): def get_accuracy(self, test_loader, *args, **kwargs): acc = 0 sample_num = 0 tokenizer = getTokenizer() self.to_eval_mode() pred_txt = [] true_txt = [] res = [] with torch.no_grad(): pbar = tqdm.tqdm(enumerate(test_loader), total=len(test_loader), dynamic_ncols=True, leave=False) for batch_index, (x, _) in pbar: if len(x) == 0: continue # if batch_index > 10: # break for k, v in x.items(): if isinstance(v, torch.Tensor): x[k] = v.to(self.device) # input_ids = [] inputlen = x['len'] y = x['labels'] x['labels'] = None outputs = self.models_dict['main'].generate(x, pad_id=tokenizer.pad_token_id) for i, op in enumerate(outputs): op = op.tolist() op = list(filter(lambda x: x != tokenizer.pad_token_id, op)) txt = tokenizer.decode(op) txt = txt.replace(tokenizer.pad_token, "") res.append(txt) txt = txt[inputlen[i]:] pred_txt.append(nltk.word_tokenize(txt)) for tp in y: true_txt.append(nltk.word_tokenize(tokenizer.decode(tp).replace(tokenizer.pad_token, ''))) # pred = F.softmax(output, dim=1).argmax(dim=1) # correct = torch.eq(pred, y).sum().item() # acc += correct sample_num += len(y) # pbar.set_description(f'cur_batch_total: {len(y)}, cur_batch_correct: {correct}, ' # f'cur_batch_acc: {(correct / len(y)):.4f}') json.dump(res, open("./gpt_generation.json", "w")) smooth = SmoothingFunction() score = 0. for pred, true in zip(pred_txt, true_txt): score += sentence_bleu([true], pred, weights=(0.25, 0.25, 0.25, 0.25), smoothing_function=smooth.method1) score /= sample_num return score def get_elastic_dnn_util(self) -> ElasticDNNUtil: return ElasticGPTUtil() def get_fm_matched_param_of_md_param(self, md_param_name): # only between qkv.weight, norm.weight/bias self_param_name = md_param_name fm = self.models_dict['fm'] # if any([k in self_param_name for k in ['fbs', 'cls_token', 'pos_embed']]): # return None # p = get_parameter(self.models_dict['md'], self_param_name) # if p.dim() == 0: # return None # elif p.dim() == 1 and 'norm' in self_param_name and 'weight' in self_param_name: # return get_parameter(fm, self_param_name) if any([k in self_param_name for k in ['fbs', 'wte', 'wpe']]): return None p = get_parameter(self.models_dict['md'], self_param_name) if p.dim() == 0: return None # elif p.dim() == 1 and 'layernorm' in self_param_name and 'weight' in self_param_name: # return get_parameter(fm, self_param_name) # 1. xx.qkv.to_qkv.yy to xx.qkv.qkv.aa and xx.qkv.abs.zz # if 'qkv.weight' in self_param_name: # ss = self_param_name.split('.') # fm_qkv_name = '.'.join(ss[0: -1]) + '.qkv' # fm_qkv = get_module(fm, fm_qkv_name) # fm_abs_name = '.'.join(ss[0: -1]) + '.abs' # fm_abs = get_module(fm, fm_abs_name) # # NOTE: unrecoverable operation! multiply LoRA parameters to allow it being updated in update_fm_param() # # TODO: if fm will be used for inference, _mul_lora_weight will not be applied! # if not hasattr(fm_abs, '_mul_lora_weight'): # logger.debug(f'set _mul_lora_weight in {fm_abs_name}') # setattr(fm_abs, '_mul_lora_weight', # nn.Parameter(torch.cat([(_abs[0].weight.T @ _abs[1].weight.T).T for _abs in fm_abs], dim=0))) # return torch.cat([ # fm_qkv.weight.data, # task-agnositc params # fm_abs._mul_lora_weight.data # task-specific params (LoRA) # ], dim=0) # # elif 'to_qkv.bias' in self_param_name: # # ss = self_param_name.split('.') # # fm_qkv_name = '.'.join(ss[0: -2]) + '.qkv.bias' # # return get_parameter(fm, fm_qkv_name) # elif 'mlp.fc1' in self_param_name and 'weight' in self_param_name: # fm_param_name = self_param_name.replace('.linear', '') # return get_parameter(fm, fm_param_name) # elif 'mlp.fc2' in self_param_name and 'weight' in self_param_name: # fm_param_name = self_param_name # return get_parameter(fm, fm_param_name) # else: # # return get_parameter(fm, self_param_name) # return None if ('q_proj' in self_param_name or 'k_proj' in self_param_name or \ 'v_proj' in self_param_name) and ('weight' in self_param_name): ss = self_param_name.split('.') fm_qkv_name = '.'.join(ss[0: -1]) + '.fc' fm_qkv = get_module(fm, fm_qkv_name) fm_abs_name = '.'.join(ss[0: -1]) + '.ab' fm_abs = get_module(fm, fm_abs_name) if not hasattr(fm_abs, '_mul_lora_weight'): logger.debug(f'set _mul_lora_weight in {fm_abs_name}') setattr(fm_abs, '_mul_lora_weight', nn.Parameter(fm_abs[1].weight @ fm_abs[0].weight)) return torch.cat([ fm_qkv.weight.data, # task-agnositc params fm_abs._mul_lora_weight.data # task-specific params (LoRA) ], dim=0) elif ('q_proj' in self_param_name or 'k_proj' in self_param_name or \ 'v_proj' in self_param_name) and ('bias' in self_param_name): ss = self_param_name.split('.') fm_qkv_name = '.'.join(ss[0: -1]) + '.fc.bias' return get_parameter(fm, fm_qkv_name) elif 'mlp.c_fc' in self_param_name and 'weight' in self_param_name: fm_param_name = self_param_name.replace('.linear', '') return get_parameter(fm, fm_param_name) elif 'mlp.c_fc' in self_param_name and 'bias' in self_param_name: fm_param_name = self_param_name.replace('.linear', '') return get_parameter(fm, fm_param_name) # elif 'mlp.fc2' in self_param_name and 'weight' in self_param_name: # fm_param_name = self_param_name.replace('.linear', '') # return get_parameter(fm, fm_param_name) else: #return get_parameter(fm, self_param_name) return None def update_fm_param(self, md_param_name, cal_new_fm_param_by_md_param): if not (('q_proj' in md_param_name or 'k_proj' in md_param_name or \ 'v_proj' in md_param_name) and ('weight' in md_param_name)): matched_fm_param_ref = self.get_fm_matched_param_of_md_param(md_param_name) matched_fm_param_ref.copy_(cal_new_fm_param_by_md_param) else: new_fm_attn_weight, new_fm_lora_weight = torch.chunk(cal_new_fm_param_by_md_param, 2, 0) ss = md_param_name.split('.') fm = self.models_dict['fm'] # update task-agnostic parameters fm_qkv_name = '.'.join(ss[0: -1]) + '.fc' fm_qkv = get_module(fm, fm_qkv_name) fm_qkv.weight.data.copy_(new_fm_attn_weight) # update task-specific parameters fm_abs_name = '.'.join(ss[0: -1]) + '.ab' fm_abs = get_module(fm, fm_abs_name) fm_abs._mul_lora_weight.data.copy_(new_fm_lora_weight) # TODO: this will not be applied in inference! def get_md_matched_param_of_fm_param(self, fm_param_name): return super().get_md_matched_param_of_fm_param(fm_param_name) def get_md_matched_param_of_sd_param(self, sd_param_name): # raise NotImplementedError # only between qkv.weight, norm.weight/bias self_param_name = sd_param_name md = self.models_dict['md'] if any([k in self_param_name for k in ['fbs', 'wte', 'wpe']]): return None p = get_parameter(self.models_dict['sd'], self_param_name) if p.dim() == 0: return None elif p.dim() == 1 and ('LayerNorm' in self_param_name or 'ln' in self_param_name) and 'weight' in self_param_name: return get_parameter(md, self_param_name) # 1. xx.qkv.to_qkv.yy to xx.qkv.qkv.aa and xx.qkv.abs.zz if ('q_proj' in self_param_name or 'k_proj' in self_param_name or \ 'v_proj' in self_param_name) and ('weight' in self_param_name): return get_parameter(md, self_param_name) # NOTE: no fbs in qkv! # elif 'to_qkv.bias' in self_param_name: # ss = self_param_name.split('.') # fm_qkv_name = '.'.join(ss[0: -2]) + '.qkv.bias' # return get_parameter(fm, fm_qkv_name) elif 'mlp.c_fc.0.weight' in self_param_name: fm_param_name = '.'.join(self_param_name.split('.')[0: -2]) + '.linear.weight' return get_parameter(md, fm_param_name) elif 'mlp.c_fc.0.bias' in self_param_name: fm_param_name = '.'.join(self_param_name.split('.')[0: -2]) + '.linear.bias' return get_parameter(md, fm_param_name) elif 'mlp.c_proj' in self_param_name and 'weight' in self_param_name: fm_param_name = self_param_name return get_parameter(md, fm_param_name) elif 'static_channel_attention' not in self_param_name: return get_parameter(md, self_param_name) # return None def get_task_head_params(self): head = get_module(self.models_dict['sd'], 'classifier') return list(head.parameters()) class TxtgenOnlineFeatAlignModel(OnlineFeatAlignModel): def get_trained_params(self): # TODO: elastic fm only train a part of params #qkv_and_norm_params = [p for n, p in self.models_dict['main'].named_parameters() if 'attention.attention.projection_query' in n or 'attention.attention.projection_key' in n or 'attention.attention.projection_value' in n or 'intermediate.dense' in n or 'output.dense' in n] qkv_and_norm_params = [p for n, p in self.models_dict['main'].named_parameters()] return qkv_and_norm_params def get_feature_hook(self) -> LayerActivation2: return LayerActivation2(get_module(self.models_dict['main'], 'model.lm_head')) def forward_to_get_task_loss(self, x, y): losses = self.infer(x) # print(losses) return losses def get_mmd_loss(self, f1, f2): common_shape = min(f1.shape[0], f2.shape[0]) f1 = f1.view(f1.shape[0], -1) f2 = f2.view(f2.shape[0], -1) f1 = f1[:common_shape,:] f2 = f2[:common_shape,:] return mmd_rbf(f1, f2) def infer(self, x, *args, **kwargs): return self.models_dict['main'](**x) def get_accuracy(self, test_loader, *args, **kwargs): acc = 0 sample_num = 0 tokenizer = getTokenizer() self.to_eval_mode() pred_txt = [] true_txt = [] res = [] with torch.no_grad(): pbar = tqdm.tqdm(enumerate(test_loader), total=len(test_loader), dynamic_ncols=True, leave=False) for batch_index, (x, _) in pbar: if len(x) == 0: continue # if batch_index > 10: # break for k, v in x.items(): if isinstance(v, torch.Tensor): x[k] = v.to(self.device) # input_ids = [] inputlen = x['len'] y = x['labels'] x['labels'] = None outputs = self.models_dict['main'].generate(x, pad_id=tokenizer.pad_token_id) for i, op in enumerate(outputs): op = op.tolist() op = list(filter(lambda x: x != tokenizer.pad_token_id, op)) txt = tokenizer.decode(op) txt = txt.replace(tokenizer.pad_token, "") res.append(txt) txt = txt[inputlen[i]:] pred_txt.append(nltk.word_tokenize(txt)) for tp in y: true_txt.append(nltk.word_tokenize(tokenizer.decode(tp).replace(tokenizer.pad_token, ''))) # pred = F.softmax(output, dim=1).argmax(dim=1) # correct = torch.eq(pred, y).sum().item() # acc += correct sample_num += len(y) # pbar.set_description(f'cur_batch_total: {len(y)}, cur_batch_correct: {correct}, ' # f'cur_batch_acc: {(correct / len(y)):.4f}') json.dump(res, open("./gpt_generation.json", "w")) smooth = SmoothingFunction() score = 0. for pred, true in zip(pred_txt, true_txt): score += sentence_bleu([true], pred, weights=(0.25, 0.25, 0.25, 0.25), smoothing_function=smooth.method1) score /= sample_num return score #from new_impl.cv.model import ElasticDNN_ClsOnlineModel elasticfm_model = ElasticDNN_TxtgenOnlineModel('gen', init_online_model( 'new_impl/nlp/gpt-neo/text_generation/results/gen_md_w_fbs_index.py/20231222/999995-003118-results/models/fm_best.pt', 'new_impl/nlp/gpt-neo/text_generation/results/gen_md_w_fbs_index.py/20231222/999995-003118-results/models/md_best.pt', 'gen', __file__ ), device, { 'md_to_fm_alpha': 0.01, 'fm_to_md_alpha': 0.1 }) da_alg = FeatAlignAlg from utils.dl.common.lr_scheduler import get_linear_schedule_with_warmup #from new_impl.cv.model import ClsOnlineFeatAlignModel da_model = TxtgenOnlineFeatAlignModel da_alg_hyp = { 'Medicine_task': { 'train_batch_size': 2, 'val_batch_size': 1, 'num_workers': 2, 'optimizer': 'AdamW', 'optimizer_args': {'lr': 1e-6, 'betas': [0.9, 0.999], 'weight_decay': 0.01}, 'scheduler': '', 'scheduler_args': {}, 'num_iters': 1000, 'val_freq': 200, 'sd_sparsity':0.3, 'feat_align_loss_weight': 1.0, }, 'Law_task': { 'train_batch_size': 2, 'val_batch_size': 1, 'num_workers': 2, 'optimizer': 'AdamW', 'optimizer_args': {'lr': 1e-6, 'betas': [0.9, 0.999], 'weight_decay': 0.01}, 'scheduler': '', 'scheduler_args': {}, 'num_iters': 1000, 'val_freq': 200, 'sd_sparsity':0.3, 'feat_align_loss_weight': 1.0, }, } elasticfm_da( [app_name], [scenario], [elasticfm_model], [da_alg], [da_alg_hyp], [da_model], device, settings, __file__, "results", collate_fn=collate_fn )