*Lab: Finetuning

Contents

18. *Lab: Finetuning#

import math
from typing import List, Optional, Tuple, Union
import os
import torch
import torch.nn.functional as F
import torch.utils.checkpoint
from torch import nn
import json
import transformers
from transformers import AutoTokenizer, AutoModelForCausalLM
from functools import partial

18.1. Decoder for language modeling#

Decoder with language modeling is the previous stacked decoder layer plus a linear layer as language prediction head. The langauge prediciton head linearly transforms the hidden state into the logits distributed over the vocabulary space.

class LlamaForCausalLM(nn.Module):
    #_tied_weights_keys = ["lm_head.weight"]

    def __init__(self, config):
        super().__init__()
        self.model = LlamaModel(config)
        self.vocab_size = config.vocab_size
        self.lm_head = nn.Linear(config.hidden_size, config.vocab_size, bias=False)

    def forward(
        self,
        input_ids: torch.LongTensor = None,
    ):
        # decoder outputs consists of (dec_features, layer_state, dec_hidden, dec_attn)
        outputs = self.model(
            input_ids=input_ids,
        )

        hidden_states = outputs

        logits = self.lm_head(hidden_states)

        return logits

18.2. Data#

from torch.utils.data import Dataset, DataLoader

def format_input(entry):
    instruction_text = (
        f"Below is an instruction that describes a task. "
        f"Write a response that appropriately completes the request."
        f"\n\n### Instruction:\n{entry['instruction']}"
    )

    input_text = f"\n\n### Input:\n{entry['input']}" if entry["input"] else ""

    return instruction_text + input_text + "\n\n### Response:\n"

class InstructionDataset(Dataset):
    def __init__(self, data):
        super().__init__()
        self.prompt_with_completions = []
        self.completions = []
        
        for entry in data:
            
            instruction_plus_input = format_input(entry)
            completion = entry['output']
            
            self.prompt_with_completions.append(instruction_plus_input + completion)
            self.completions.append(completion)
        
            
    def __len__(self):
        return len(self.prompt_with_completions)
    
    def __getitem__(self, idx):
        return self.prompt_with_completions[idx], self.completions[idx]

def custom_collate_fn(batch, tokenizer, ignore_idx=-100):
    
    prompt_with_completions, completions = zip(*batch)
    
    padded = tokenizer(list(prompt_with_completions), padding='longest', truncation=True, return_tensors='pt')
    padded = padded['input_ids']
    # targets are shifted by 1
    # The last token in padded is not used in inputs as there is no corresponding target.
    inputs = padded[:,:-1]
    targets = padded[:,1:]
    
    mask = (targets == tokenizer.pad_token_id)
    
    targets = targets.masked_fill(mask, ignore_idx)
    
    return inputs, targets
    
     
def create_data_loader(data, tokenizer, batch_size=4, max_length=256, 
                       stride=128, shuffle=True, drop_last=True, num_workers=0):
    
    dataset = InstructionDataset(data=data)
    
    collate_fn = partial(custom_collate_fn, tokenizer=tokenizer, ignore_idx=-100)
    
    data_loader = DataLoader(dataset, 
                             batch_size=batch_size,
                             shuffle=shuffle,
                             drop_last=drop_last,
                             num_workers=num_workers,
                             collate_fn=collate_fn)
    
    return data_loader

def read_text_data(file_path, url):
    import urllib
    if not os.path.exists(file_path):
        with urllib.request.urlopen(url) as response:
            text_data = response.read().decode('utf-8')
        with open(file_path, "w", encoding="utf-8") as file:
            file.write(text_data)
    else:
        with open(file_path, "r", encoding="utf-8") as file:
            text_data = file.read()

    with open(file_path, "r", encoding="utf-8") as file:
        data = json.load(file)
          
    return data

import transformers
from transformers import AutoTokenizer
def test_data_component():
    file_path = "instruction-data2.json"
    url = (
        "https://raw.githubusercontent.com/tatsu-lab/stanford_alpaca/refs/heads/main/alpaca_data.json"        
    )

    text_data = read_text_data(file_path, url)
    tokenizer_name = "Qwen/Qwen2.5-0.5B"
    tokenizer = AutoTokenizer.from_pretrained(tokenizer_name)
    #tokenizer = tiktoken.get_encoding('gpt2')
    train_loader = create_data_loader(data=text_data, 
                                      tokenizer=tokenizer)
    for batch in train_loader:
        print(batch)
        break
    
test_data_component()

(tensor([[ 38214,    374,    458,   7600,    429,  16555,    264,   3383,     13,
           9645,    264,   2033,    429,  34901,  44595,    279,   1681,    382,
          14374,  29051,    510,  31115,    264,  15908,  40158,   6524,    382,
          14374,   5571,    510,     40,   1079,  14589,    369,    537,  33094,
            847,  16319,    389,    882,    382,  14374,   5949,    510,  30665,
            508,    675,  49088,     40,  36879,    369,  33094,    847,  16319,
           3309,     13,    358,   3535,    429,  33094,    975,    389,    882,
            374,   1376,    311,    697,   6950,    304,    752,     13,    358,
           1079,   8480,    369,    279,   7626,    323,    358,   1896,   2480,
          38142,    369,    432,     13,    358,  58283,  22231,    894,  60009,
           8881,    553,    847,   6168,    382,     40,  15440,    429,    358,
            686,   1896,    678,   5871,   7354,    311,   5978,    429,   1741,
            458,  10455,   1558,    537,   3537,   1549,     13,   9063,  27378,
             11,    358,    686,    653,    847,   1850,    311,   1281,    705,
            369,    847,  20643,    382,     40,   1401,   4637,    311,    279,
           6638,    315,  49103,    697,   6950,    304,    752,    382,     50,
          86091,    345,     58,   7771,   3988],
        [ 38214,    374,    458,   7600,    429,  16555,    264,   3383,     13,
           9645,    264,   2033,    429,  34901,  44595,    279,   1681,    382,
          14374,  29051,    510,  60424,   2326,  10295,    315,  65060,    315,
            220,     21,    382,  14374,   5949,    510,  19641,  10295,    315,
          65060,    315,    220,     21,    525,    220,     21,     11,    220,
             16,     17,     11,    323,    220,     16,     23,     13, 151643,
         151643, 151643, 151643, 151643, 151643, 151643, 151643, 151643, 151643,
         151643, 151643, 151643, 151643, 151643, 151643, 151643, 151643, 151643,
         151643, 151643, 151643, 151643, 151643, 151643, 151643, 151643, 151643,
         151643, 151643, 151643, 151643, 151643, 151643, 151643, 151643, 151643,
         151643, 151643, 151643, 151643, 151643, 151643, 151643, 151643, 151643,
         151643, 151643, 151643, 151643, 151643, 151643, 151643, 151643, 151643,
         151643, 151643, 151643, 151643, 151643, 151643, 151643, 151643, 151643,
         151643, 151643, 151643, 151643, 151643, 151643, 151643, 151643, 151643,
         151643, 151643, 151643, 151643, 151643, 151643, 151643, 151643, 151643,
         151643, 151643, 151643, 151643, 151643, 151643, 151643, 151643, 151643,
         151643, 151643, 151643, 151643, 151643],
        [ 38214,    374,    458,   7600,    429,  16555,    264,   3383,     13,
           9645,    264,   2033,    429,  34901,  44595,    279,   1681,    382,
          14374,  29051,    510,  22550,   1493,   5837,   4092,    311,    862,
          42807,  51749,   5643,    271,  14374,   5571,    510,  24347,     11,
           9856,     11,   9625,     11,   6747,    271,  14374,   5949,    510,
          24347,     11,   9856,     11,   6747,     11,   9625, 151643, 151643,
         151643, 151643, 151643, 151643, 151643, 151643, 151643, 151643, 151643,
         151643, 151643, 151643, 151643, 151643, 151643, 151643, 151643, 151643,
         151643, 151643, 151643, 151643, 151643, 151643, 151643, 151643, 151643,
         151643, 151643, 151643, 151643, 151643, 151643, 151643, 151643, 151643,
         151643, 151643, 151643, 151643, 151643, 151643, 151643, 151643, 151643,
         151643, 151643, 151643, 151643, 151643, 151643, 151643, 151643, 151643,
         151643, 151643, 151643, 151643, 151643, 151643, 151643, 151643, 151643,
         151643, 151643, 151643, 151643, 151643, 151643, 151643, 151643, 151643,
         151643, 151643, 151643, 151643, 151643, 151643, 151643, 151643, 151643,
         151643, 151643, 151643, 151643, 151643, 151643, 151643, 151643, 151643,
         151643, 151643, 151643, 151643, 151643],
        [ 38214,    374,    458,   7600,    429,  16555,    264,   3383,     13,
           9645,    264,   2033,    429,  34901,  44595,    279,   1681,    382,
          14374,  29051,    510,   4340,   2310,    374,    279,  88575,    315,
          31392,   1939,  14374,   5949,    510,    785,  88575,    315,  31392,
            374,    916,    220,     16,     18,     19,   1635,   2310,     11,
           3432,   1012,   8145,    304,   6527,    220,     16,     23,     23,
             21,     13, 151643, 151643, 151643, 151643, 151643, 151643, 151643,
         151643, 151643, 151643, 151643, 151643, 151643, 151643, 151643, 151643,
         151643, 151643, 151643, 151643, 151643, 151643, 151643, 151643, 151643,
         151643, 151643, 151643, 151643, 151643, 151643, 151643, 151643, 151643,
         151643, 151643, 151643, 151643, 151643, 151643, 151643, 151643, 151643,
         151643, 151643, 151643, 151643, 151643, 151643, 151643, 151643, 151643,
         151643, 151643, 151643, 151643, 151643, 151643, 151643, 151643, 151643,
         151643, 151643, 151643, 151643, 151643, 151643, 151643, 151643, 151643,
         151643, 151643, 151643, 151643, 151643, 151643, 151643, 151643, 151643,
         151643, 151643, 151643, 151643, 151643, 151643, 151643, 151643, 151643,
         151643, 151643, 151643, 151643, 151643]]), tensor([[  374,   458,  7600,   429, 16555,   264,  3383,    13,  9645,   264,
          2033,   429, 34901, 44595,   279,  1681,   382, 14374, 29051,   510,
         31115,   264, 15908, 40158,  6524,   382, 14374,  5571,   510,    40,
          1079, 14589,   369,   537, 33094,   847, 16319,   389,   882,   382,
         14374,  5949,   510, 30665,   508,   675, 49088,    40, 36879,   369,
         33094,   847, 16319,  3309,    13,   358,  3535,   429, 33094,   975,
           389,   882,   374,  1376,   311,   697,  6950,   304,   752,    13,
           358,  1079,  8480,   369,   279,  7626,   323,   358,  1896,  2480,
         38142,   369,   432,    13,   358, 58283, 22231,   894, 60009,  8881,
           553,   847,  6168,   382,    40, 15440,   429,   358,   686,  1896,
           678,  5871,  7354,   311,  5978,   429,  1741,   458, 10455,  1558,
           537,  3537,  1549,    13,  9063, 27378,    11,   358,   686,   653,
           847,  1850,   311,  1281,   705,   369,   847, 20643,   382,    40,
          1401,  4637,   311,   279,  6638,   315, 49103,   697,  6950,   304,
           752,   382,    50, 86091,   345,    58,  7771,  3988,    60],
        [  374,   458,  7600,   429, 16555,   264,  3383,    13,  9645,   264,
          2033,   429, 34901, 44595,   279,  1681,   382, 14374, 29051,   510,
         60424,  2326, 10295,   315, 65060,   315,   220,    21,   382, 14374,
          5949,   510, 19641, 10295,   315, 65060,   315,   220,    21,   525,
           220,    21,    11,   220,    16,    17,    11,   323,   220,    16,
            23,    13,  -100,  -100,  -100,  -100,  -100,  -100,  -100,  -100,
          -100,  -100,  -100,  -100,  -100,  -100,  -100,  -100,  -100,  -100,
          -100,  -100,  -100,  -100,  -100,  -100,  -100,  -100,  -100,  -100,
          -100,  -100,  -100,  -100,  -100,  -100,  -100,  -100,  -100,  -100,
          -100,  -100,  -100,  -100,  -100,  -100,  -100,  -100,  -100,  -100,
          -100,  -100,  -100,  -100,  -100,  -100,  -100,  -100,  -100,  -100,
          -100,  -100,  -100,  -100,  -100,  -100,  -100,  -100,  -100,  -100,
          -100,  -100,  -100,  -100,  -100,  -100,  -100,  -100,  -100,  -100,
          -100,  -100,  -100,  -100,  -100,  -100,  -100,  -100,  -100,  -100,
          -100,  -100,  -100,  -100,  -100,  -100,  -100,  -100,  -100],
        [  374,   458,  7600,   429, 16555,   264,  3383,    13,  9645,   264,
          2033,   429, 34901, 44595,   279,  1681,   382, 14374, 29051,   510,
         22550,  1493,  5837,  4092,   311,   862, 42807, 51749,  5643,   271,
         14374,  5571,   510, 24347,    11,  9856,    11,  9625,    11,  6747,
           271, 14374,  5949,   510, 24347,    11,  9856,    11,  6747,    11,
          9625,  -100,  -100,  -100,  -100,  -100,  -100,  -100,  -100,  -100,
          -100,  -100,  -100,  -100,  -100,  -100,  -100,  -100,  -100,  -100,
          -100,  -100,  -100,  -100,  -100,  -100,  -100,  -100,  -100,  -100,
          -100,  -100,  -100,  -100,  -100,  -100,  -100,  -100,  -100,  -100,
          -100,  -100,  -100,  -100,  -100,  -100,  -100,  -100,  -100,  -100,
          -100,  -100,  -100,  -100,  -100,  -100,  -100,  -100,  -100,  -100,
          -100,  -100,  -100,  -100,  -100,  -100,  -100,  -100,  -100,  -100,
          -100,  -100,  -100,  -100,  -100,  -100,  -100,  -100,  -100,  -100,
          -100,  -100,  -100,  -100,  -100,  -100,  -100,  -100,  -100,  -100,
          -100,  -100,  -100,  -100,  -100,  -100,  -100,  -100,  -100],
        [  374,   458,  7600,   429, 16555,   264,  3383,    13,  9645,   264,
          2033,   429, 34901, 44595,   279,  1681,   382, 14374, 29051,   510,
          4340,  2310,   374,   279, 88575,   315, 31392,  1939, 14374,  5949,
           510,   785, 88575,   315, 31392,   374,   916,   220,    16,    18,
            19,  1635,  2310,    11,  3432,  1012,  8145,   304,  6527,   220,
            16,    23,    23,    21,    13,  -100,  -100,  -100,  -100,  -100,
          -100,  -100,  -100,  -100,  -100,  -100,  -100,  -100,  -100,  -100,
          -100,  -100,  -100,  -100,  -100,  -100,  -100,  -100,  -100,  -100,
          -100,  -100,  -100,  -100,  -100,  -100,  -100,  -100,  -100,  -100,
          -100,  -100,  -100,  -100,  -100,  -100,  -100,  -100,  -100,  -100,
          -100,  -100,  -100,  -100,  -100,  -100,  -100,  -100,  -100,  -100,
          -100,  -100,  -100,  -100,  -100,  -100,  -100,  -100,  -100,  -100,
          -100,  -100,  -100,  -100,  -100,  -100,  -100,  -100,  -100,  -100,
          -100,  -100,  -100,  -100,  -100,  -100,  -100,  -100,  -100,  -100,
          -100,  -100,  -100,  -100,  -100,  -100,  -100,  -100,  -100]]))

18.3. Training#

def compute_batch_loss(input_batch, target_batch, model, device):
    input_batch, target_batch = input_batch.to(device), target_batch.to(device)
    logits = model(input_batch)
    
    flat_targets = target_batch.flatten() 
    flat_logits = logits.flatten(0, 1)# flatten the first two dimensions 
    loss = F.cross_entropy(flat_logits, flat_targets) # tokens with ignore idx will not contribute to loss 
    return loss

def train_model_epoch(model, 
                train_loader,
                optimizer,
                device,
                num_epochs):
    
    train_losses, val_losses, track_token_seen = [],[],[]
    tokens_seen = 0
    global_steps = -1
    
    for epoch in range(num_epochs):
        model.train()
        
        for input_batch, target_batch in train_loader:
            optimizer.zero_grad()
            loss = compute_batch_loss(input_batch, target_batch, model, device)
            loss.backward()
            optimizer.step()
            tokens_seen += input_batch.numel()
            global_steps += 1
            train_losses.append(loss.detach().item())
            print(train_losses[-1])
        
    return train_losses, model

def train_main(model_config, train_settings):
    
    torch.manual_seed(train_settings.seed)
    device = torch.device('cuda') if torch.cuda.is_available() else torch.device('cpu')
    
    text_data = read_text_data(train_settings.file_path, train_settings.url)
            
    model = LlamaForCausalLM(config=model_config)
    model.to(device)
    optimizer = torch.optim.AdamW(model.parameters(),
                                  lr=train_settings.learning_rate,
                                  weight_decay=train_settings.weight_decay)
    
    # set up dataloader
    
    
    tokenizer = AutoTokenizer.from_pretrained(train_settings.pretrained_model_name)
    train_loader = create_data_loader(data=text_data, 
                                      tokenizer=tokenizer)
    
    
    train_loader = create_data_loader(data=text_data,
                                      tokenizer=tokenizer,
                                      batch_size=train_settings.batch_size,
                                      drop_last=True,
                                      shuffle=True,
                                        num_workers=0
    )
        
    train_losses, model = train_model_epoch(model=model,
                train_loader=train_loader,
                optimizer=optimizer,
                num_epochs=train_settings.num_epochs,
                device=device)
    
    return train_losses, model
    

if __name__ == '__main__':
    
    model_config = {
        "attention_dropout": 0.0,
        "bos_token_id": 151643,
        "eos_token_id": 151643,
        "pad_token_id": 151643,
        "hidden_act": "silu",
        "hidden_size": 896,
        "initializer_range": 0.02,
        "intermediate_size": 4864,
        "max_position_embeddings": 32768,
        "max_window_layers": 24,
        "model_type": "qwen2",
        "num_attention_heads": 14,
        "num_hidden_layers": 24,
        "num_key_value_heads": 2,
        "rms_norm_eps": 1e-06,
        "rope_theta": 1000000.0,
        "tie_word_embeddings": True,
        "torch_dtype": "bfloat16",
        "transformers_version": "4.47.1",
        "use_cache": True,
        "use_mrope": False,
        "vocab_size": 151936,
        "qkv_bias": True,
        "o_bias": False,
        "mlp_bias": False
    }

    model_config = OmegaConf.create(model_config)
    train_settings = {
        "pretrained_model_name": "Qwen/Qwen2.5-0.5B",
        "learning_rate": 5e-6,
        "num_epochs": 10,
        "batch_size": 4,
        "weight_decay": 0.1,
        "stride": 128,
        "seed": 1,
        "file_path":"./instruction_data/instruction-data2.json",
        "url":"https://raw.githubusercontent.com/tatsu-lab/stanford_alpaca/refs/heads/main/alpaca_data.json"
    }
    
    train_settings = OmegaConf.create(train_settings)
    
    # train model
    train_losses, model = train_main(model_config=model_config,
                       train_settings=train_settings)
    
    print(train_losses)

114983558654785
961651802062988
927282333374023
826778411865234
707758903503418
793996810913086
565061569213867
341635704040527
44568920135498
51699161529541
538795471191406
205238342285156
476774215698242
2799654006958
608583450317383
964029312133789
31577205657959
75229549407959
765582084655762
845158576965332
3082857131958
781938552856445
183990478515625
394356727600098
934983253479004
947456359863281
557378768920898
630148887634277
455920219421387
409931182861328
108236312866211
182458877563477
501925468444824
636826515197754
441350936889648
442732810974121
65182113647461
005826950073242
056744575500488
071742057800293
719274520874023
539457321166992
109013557434082
988809585571289
503620147705078
56149673461914
963241577148438
929883003234863
429275512695312
508111953735352
378183364868164
728569984436035
021723747253418
781039237976074
219264030456543
941544532775879
30162525177002
398658752441406
79794979095459
475982666015625
320345878601074
690967559814453
79287338256836
732223510742188
572741508483887
09702205657959
091935157775879
1818265914917
378776550292969
197397232055664
443197250366211
796965599060059
650116920471191
56566047668457
037643432617188
316316604614258
480766296386719
105834007263184
8174004554748535
75154972076416
871994972229004
084001064300537
267117977142334
413370132446289
928682327270508
569966316223145
723995208740234
712878227233887
335203170776367
576348304748535
202093124389648
326104164123535
715723037719727
051654815673828
702801704406738
767918109893799
44614315032959
499092102050781
835850715637207
17809009552002
459793090820312
364670753479004
678229808807373
883283615112305
294384956359863
772454738616943
872432708740234
260662078857422
852313041687012
336148262023926
217145919799805
837427139282227
094305515289307
249174118041992
225741386413574
286722183227539
037725448608398
328179359436035
948586463928223
679088592529297
223731994628906
198479652404785
246973991394043
424270153045654
886054039001465
324112892150879
785794258117676
4138765335083
507905960083008
868498802185059
550424575805664
014437675476074
699291229248047
966353416442871
073481559753418
768507957458496
203961372375488
526586055755615
272527694702148
836536407470703
6048583984375
950989246368408
622305870056152
43606948852539
453845024108887
930305480957031
952410697937012
807299613952637
922638893127441
14686393737793
115427494049072
830827236175537
790492057800293
53681755065918
681835651397705
839694976806641
047902584075928
7306365966796875
827072620391846
690933227539062
336493492126465
476668834686279
079100608825684
954265594482422
466475486755371
256246566772461
341320037841797
160398006439209
395797729492188
977738380432129
152826309204102
279878616333008
9665141105651855
751248836517334
755195617675781
081643104553223
407286643981934
913785934448242
948304653167725
02336311340332
258242607116699
707741737365723
176688194274902
005913734436035
04478645324707
365195274353027
96026611328125
431405067443848
723588466644287
396443367004395
315401077270508
87647533416748
090518951416016
729640483856201
5797553062438965
677322864532471
530302047729492
831056118011475
948013305664062
121613502502441
95388650894165
317086696624756
587782382965088
112850666046143
076551914215088
187629699707031
603550434112549
574467658996582
634267807006836
89480447769165
144584655761719
720107078552246
302224636077881
600544452667236
489965438842773
439957618713379
317536354064941
296192169189453
806612491607666
755681991577148
698017597198486
634511470794678
9502949714660645
371891021728516
99277925491333
765937805175781
635364055633545
691554069519043
497350692749023
27408504486084
323247909545898
470743179321289
872669696807861
903525352478027
573783874511719
304293155670166
692304611206055
821817398071289
150716781616211
094555854797363
496853351593018
204647064208984
9287333488464355
656194686889648
684568405151367
101051330566406
001102447509766
5603532791137695
81414794921875
483189582824707
4107441902160645
123326778411865
163662910461426
600272178649902
92359733581543
966070652008057
676620960235596
071258068084717
694700717926025
357487678527832
621312618255615
154609680175781
327323913574219
954722881317139
917107582092285
6910786628723145
665449619293213
7606987953186035
196247100830078
23905086517334
622346878051758
084579467773438
474020481109619
9481024742126465
517650604248047
622471332550049
6400299072265625
146424770355225
000862121582031
315019607543945
4658331871032715
999532699584961
73932409286499
784018039703369
335762023925781
795717716217041
183656692504883
980317115783691
3821120262146
172737121582031
515994071960449
037397384643555
697264671325684
97719144821167
995906829833984
219350337982178
621826648712158
60056209564209
303577899932861
741073131561279
910447120666504
5270466804504395
173192501068115
292308330535889
583615779876709
667201995849609
151587963104248
602296352386475
106971740722656
574258804321289
1769914627075195
854578495025635
636929512023926
061482429504395
62721586227417
101871013641357
244905471801758
899890422821045
74027681350708
143261909484863
437134265899658
466038227081299
8156819343566895
31174373626709
805547714233398
186622142791748
483814716339111
559063911437988
888976573944092
138240814208984
423436641693115
507336616516113
050281047821045
500576972961426
398224353790283
023150444030762
995843887329102
583437442779541
764934539794922
284381866455078
54485559463501
271533966064453
747768878936768
4070963859558105
711648464202881
8751044273376465
873578071594238
798689842224121
163131713867188
023693561553955
624054431915283
368297576904297
57619571685791
637344837188721
854536056518555
881252765655518
129610061645508
047771453857422
673618316650391
853394985198975
695700645446777
286881446838379
918391704559326
60928201675415
443287372589111
012099266052246
712862968444824
353724956512451
589232444763184
810102939605713
120845317840576
180410385131836
988944053649902
331676006317139
220089912414551
535629749298096
000095367431641
269332408905029
273829936981201
7589945793151855
78818941116333
716200351715088
104673862457275
209777355194092
009293556213379
337612152099609
183322906494141
3463053703308105
53585433959961
489135265350342
748711585998535
630690574645996
174975395202637
681833744049072
685907363891602
3267621994018555
549046039581299
512974739074707
26035213470459
584530353546143
311887741088867
779975414276123
518989562988281
98706579208374
305261135101318
529804706573486
763954162597656
788674831390381
4324140548706055
752645969390869
385810375213623
1607208251953125
199484825134277
8876633644104
428523063659668
391116142272949
399779796600342
628757476806641
763996601104736
302529811859131
880929946899414
014496803283691
598532199859619
9200592041015625
126089096069336
731667518615723
363903999328613
017136573791504
837329864501953
472388744354248
30039644241333
211876392364502
566425323486328
906378746032715
059329509735107
300551414489746
950073719024658
633367538452148
883755207061768
330907344818115
959106922149658
559053421020508
327077388763428
252851486206055
105184078216553
396198272705078
692835807800293
1359076499938965
052042007446289
576384544372559
252415657043457
510921478271484
675665378570557
823562145233154
835978031158447
80307149887085
005776882171631
290078163146973
8772687911987305
2689290046691895
407567024230957
917119979858398
427750110626221
4736328125
155660629272461
854784965515137
569583892822266
052281379699707
647436618804932
95413064956665
043063640594482
3223795890808105
029530048370361
009037494659424
369438171386719
471277713775635
596868991851807
728212833404541
57578182220459
205798625946045
723975658416748
969112873077393
523230075836182
158575057983398
533547401428223
217704772949219
9049072265625
263994216918945
208010673522949
861074447631836
295699119567871
602899551391602
556525230407715
597959041595459
1291351318359375
438089370727539
826323509216309
69783878326416
3540425300598145
287336826324463
87325382232666
027651309967041
817617416381836
585550308227539
752303123474121
535881042480469
355349540710449
66028356552124
829235553741455
75210428237915
757355690002441
3517656326293945
707007884979248
924886226654053
391934394836426
579033374786377
863368034362793
487936019897461
390007495880127
895957946777344
898429870605469
337359428405762
9140095710754395
807152271270752
988922595977783
189905166625977
269647121429443
8126726150512695
816813945770264
724527359008789
713352203369141
766097545623779
72987174987793
434146881103516
73057746887207
03439474105835
549989223480225
1966233253479
8368377685546875
853939533233643
491785049438477
493218898773193
554964065551758
743686199188232
60521936416626
051425457000732
890924453735352
29083251953125
18527889251709
9394731521606445
673720836639404
949153900146484
510037422180176
879708290100098
112819671630859
374730110168457
098811626434326
41429328918457
876171588897705
408968448638916
09466552734375
505105972290039
930153846740723
188701629638672
083053112030029
429739475250244
784379005432129
117002010345459
531440258026123
997178077697754
864298343658447
955168724060059
899875164031982
1008782386779785
774710655212402
519045352935791
862921714782715
601291179656982
76458215713501
934217929840088
3184404373168945
798949718475342
4431843757629395
169183731079102
487637042999268
239485263824463
50405740737915
234323978424072
967166423797607
920042514801025
90633487701416
497448921203613
810624122619629
948454856872559
56923770904541
366409778594971
629382610321045
370241641998291
956958293914795
130191326141357
562791347503662
687262058258057
979550361633301
183032512664795
493767261505127
1848344802856445
289977550506592
878566741943359
262190818786621
668609619140625
653968811035156
002604961395264
90189790725708
340113162994385
9966230392456055
680359840393066
483484745025635
6956562995910645
696816921234131
163222789764404
37417459487915
459051609039307
055540084838867
523402214050293
372015953063965
172832489013672
614804267883301
371455669403076
445786476135254
082972049713135
580028533935547
932221412658691
254146099090576
9570159912109375
014609336853027
235097408294678
052845001220703
747809410095215
098481178283691
14454984664917
109280109405518
398374557495117
213296890258789
434049606323242
272828102111816
0189290046691895
264615058898926
432180404663086
706062316894531
067690849304199
473038673400879
232213497161865
500061988830566
305816173553467
634389877319336
7206292152404785
477426528930664
866785049438477
254941940307617
022000789642334
145068168640137
892409324645996
718930721282959
42185115814209
1081318855285645
033673286437988
882725715637207
184030055999756
220764636993408
509552001953125
992305278778076
082344055175781
490474700927734
823636054992676
924805164337158
592731475830078
081788539886475
7530999183654785
421682357788086
761895179748535
41145658493042
259069442749023
91668176651001
9712347984313965
9838547706604
8265886306762695
741562366485596
678469657897949
087961673736572
12254524230957
6356282234191895
999855995178223
242163181304932
566010475158691
29323148727417
052127838134766
39547872543335
888261795043945
235403537750244
532500267028809
24505090713501
629907608032227
488248825073242
229555606842041
516720771789551
067878723144531
0529069900512695
704999923706055
138185024261475
723733425140381
492624759674072
21068000793457
799619197845459
321924686431885
703210353851318
221750259399414
257620811462402
067604064941406
78139591217041
810388565063477
10854959487915
632315635681152
442394733428955
245612621307373
078483581542969
893668174743652
5216474533081055
554015159606934
825289726257324
4750823974609375
112430095672607
616897106170654
479202747344971
688720703125
061813831329346
848109722137451
831615447998047
852490425109863
025716781616211
123955249786377
684447288513184
867308616638184
07834529876709
90164852142334
524458408355713
9056620597839355
2032294273376465
065298080444336
184986114501953
666153430938721
254053115844727
970147609710693
205985069274902
008021831512451
247925281524658
835949420928955
042531490325928
213634490966797
94294548034668
186612129211426
406094074249268
199366569519043
122321128845215
584754943847656
180679798126221
580134868621826
852870941162109
759560585021973
577573776245117
284355640411377
7807536125183105
727319717407227
722000598907471
277144908905029
57779598236084
90477180480957
326879024505615
294795036315918
40275764465332
876376628875732
609960079193115
141951084136963
104902744293213
284786701202393
333367824554443
107256889343262
362699031829834
961429595947266
0032548904418945
892342567443848
109997272491455
71601676940918
278239727020264
517978191375732
3576202392578125
091977119445801
373385906219482
846373081207275
528918743133545
745575428009033
333774089813232
838168621063232
9162068367004395
3981733322143555
436718463897705
078586578369141
785589694976807
8114094734191895
154354572296143
741503715515137
176111221313477
606869220733643
833148002624512
833350658416748
255387783050537
988438606262207
452328205108643
673221111297607
423011779785156
384125709533691
643775939941406
828341007232666
667187213897705
4618144035339355
1435980796813965
366528511047363
081357955932617
075766086578369
480937957763672
268439769744873
674612998962402
6378068923950195
315445423126221
300681114196777
9949235916137695
961142539978027
220792770385742
820557594299316
383841037750244
481714248657227
074995517730713
0024333000183105
333754062652588
863290309906006
75282621383667
2128376960754395
999240875244141
917527198791504
363485336303711
912549018859863
236627101898193
727591514587402
884681701660156
498530387878418
64660120010376
371853351593018
769754409790039
753459453582764
481202602386475
8541741371154785
7227935791015625
016483306884766
684218883514404
422640323638916
581953048706055
139718532562256
288344383239746
271264553070068
095365524291992
580507755279541
741294860839844
82970666885376
041816234588623
605682849884033
488674640655518
9869866371154785
876048564910889
183134078979492
44551944732666
580467224121094
336531639099121
30568265914917
528031826019287
766164302825928
580226898193359
539986610412598
154237747192383
221808910369873
68833065032959
206247806549072
393226623535156
483890056610107
020560264587402
415234088897705
692431449890137
725409030914307
487700462341309
484647274017334
3354811668396
387298583984375
872298240661621
732889175415039
9597320556640625
123317718505859
047027587890625
713091850280762
43989896774292
4997334480285645
09600830078125
456425189971924
081859588623047
143166542053223
537712097167969
485445499420166
451141357421875
85955286026001
603623867034912
343040943145752
895185947418213
088583946228027
178514003753662
150128364562988
811209678649902
331404685974121
5498576164245605
18892765045166
646106719970703
1795573234558105
519378185272217
144201278686523
021180152893066
035079002380371
4208574295043945
481831073760986
347463607788086
000537395477295
070650100708008
531777381896973
393858432769775
812191009521484
502859115600586
52064323425293
393649101257324
5129594802856445
0064697265625
120830535888672
94623327255249
661935806274414
367886066436768
446750640869141
837181568145752
475765228271484
495880603790283
333951950073242
231400489807129
890717029571533
500189781188965
826181888580322
315254211425781
590473175048828
152644634246826
569487571716309
952450275421143
208878993988037
6105570793151855
393913745880127
2702107429504395
821797847747803
272322654724121
007287502288818
845762729644775
0829901695251465
326286792755127
137276649475098
797081470489502
583113193511963
723069190979004
9189252853393555
666990756988525
260522842407227
943296194076538
16051721572876
418442249298096
377189636230469
632775783538818
806467056274414
3496479988098145
914324760437012
750854969024658
281500816345215
332863807678223
1956353187561035
598954200744629
23215913772583
006951332092285
892027854919434
189830780029297
273373603820801
675600528717041
774920463562012
101280689239502
426880836486816
9738078117370605
162805080413818
998691082000732
236884117126465
372688293457031
440547466278076
061509609222412
496494293212891
642920017242432
741084098815918
167695999145508
227993965148926
357344627380371
5911970138549805
26072883605957
713385581970215
094918251037598
06983757019043
055233478546143
5529890060424805
663405895233154
537608623504639
568994998931885
071442127227783
107507228851318
914765357971191
529574394226074
10886287689209
124289035797119
114251136779785
528431415557861
219285011291504
504852771759033
710177421569824
006173133850098
444087982177734
8146843910217285
098686695098877
489388465881348
107565879821777
020852088928223
529670000076294
425512790679932
462956428527832
309920787811279
3796772956848145
767960786819458
664869785308838
2069501876831055
6382832527160645
12039041519165
074522972106934
202847957611084
132767200469971
948268890380859
068182945251465
452475070953369
550425052642822
85136079788208
359533786773682
44975471496582
219789028167725
911383628845215
371190547943115
459372520446777
7417104244232178
802001953125
5048041343688965
0523810386657715
197486877441406
78045129776001
151343822479248
98590087890625
458860397338867
499148368835449
1656084060668945
8167617321014404
208345413208008
477724552154541
438752174377441
872332572937012
432761192321777
280029296875
3968892097473145
592565059661865
774909019470215
747305393218994
8931796550750732
775512218475342
37682580947876
683590888977051
429806232452393
472513198852539
522157669067383
326934814453125
349497318267822
28593635559082
402266025543213
525822639465332
730113983154297
715023994445801
692702293395996
210110664367676
010351657867432
959136009216309
635852336883545
019588947296143
4624552726745605
184566020965576
18332052230835
158417224884033
001770973205566
680325984954834
394460201263428
366771697998047
110700607299805
0587334632873535
516447067260742
91188907623291
885791301727295
2201690673828125
558029651641846
269782066345215
678136348724365
334537982940674
081141948699951
799370288848877
1246867179870605
551836967468262
251805305480957
931658744812012
085419178009033
518763065338135
023902416229248
544711589813232
4774370193481445
468654155731201
238506317138672
671813488006592
610466957092285
850610256195068
452916622161865
79330587387085
692015171051025
050431728363037
477248191833496
8953351974487305
819911003112793
681352615356445
495570659637451
924680709838867
041174411773682
597620487213135
588257312774658
310380458831787
292008399963379
303918361663818
013998031616211
741250038146973
334575653076172
401310443878174
825384616851807
175565719604492
432821750640869
833383560180664
086000442504883
312560081481934
37199068069458
275448322296143
90108060836792
0184197425842285
760083198547363
162322044372559
785065174102783
9133124351501465
1294450759887695
858340263366699
199103832244873
309408664703369
076613426208496
038869857788086
431438446044922
92586088180542
584958076477051
190706729888916
179817199707031
977041721343994
552811145782471
489042282104492
988609313964844
906666278839111
3543524742126465
681512355804443
897405624389648
640043258666992
832583427429199
539644241333008
7093892097473145
4353928565979
786230087280273
885951042175293
253942489624023
543970584869385
708553314208984
371676445007324
889729976654053
47752046585083
298480033874512
192046165466309
267175197601318
079553604125977
766860485076904
992218255996704
896338701248169
822120189666748
127451419830322
312523365020752
221405029296875
093663692474365
254385948181152
113960266113281
5514726638793945
9142746925354
505090713500977
273721694946289
132808685302734
226304531097412
9441943168640137
3724923133850098
139926910400391
861825942993164
884730815887451
888372898101807
8482837677001953
29171895980835
491201877593994
071336269378662
183904647827148
213995456695557
0411882400512695
471054553985596
093935012817383
739846706390381
727088212966919
26678991317749
275712490081787
569152355194092
9288668632507324
704917907714844
048269271850586
00111198425293
64829683303833
233859062194824
707916498184204
80978536605835
704864501953125
434628486633301
939565896987915
4315948486328125
682058334350586
8262834548950195
646474838256836
5544586181640625
918045997619629
438453674316406
848433494567871
662009239196777
134390830993652
9410641193389893
932360649108887
71079158782959
601119041442871
814231872558594
428111553192139
458085536956787
331404447555542
60315465927124
226582050323486
4721360206604
9737207889556885
586897850036621
276309013366699
122282028198242
927736759185791
7702178955078125
311150074005127
297110557556152
103277206420898
9419379234313965
372903347015381
102071762084961
7880635261535645
203221321105957
306757926940918
7493696212768555
061917304992676
281369686126709
243719577789307
889030933380127
008838176727295
624549865722656
673074245452881
151601791381836
921731472015381
423229217529297
706840991973877
387340068817139
423009395599365
6462178230285645
667232513427734
819949626922607
093469619750977
925262928009033
259034156799316
703780651092529
020669937133789
509369850158691
062737226486206
142984390258789
0959885120391846
622593879699707
393157005310059
985677719116211
704891681671143
81308650970459
951192378997803
4936649799346924
268042087554932
648794651031494
741124629974365
719524383544922
132901668548584
6371138095855713
121431827545166
394814491271973
495635032653809
504574775695801
935726165771484
019296646118164
411169052124023
481995105743408
133650779724121
87380313873291
370524883270264
176878452301025
607332706451416
8344197273254395
280559539794922
755044937133789
670968532562256
552657604217529
183326244354248
28824520111084
536715030670166
266002655029297
19411039352417
059132099151611
332251071929932
9927866458892822
1863226890563965
6914448738098145
380285739898682
6727049350738525
2017765045166016
617480754852295
1258463859558105
2450714111328125
4287636280059814
3055548667907715
532714605331421
76379919052124
456601619720459
968337297439575
906172513961792
850433349609375
110368728637695
797332763671875
18759298324585
007312297821045
3863539695739746
900402545928955
734208822250366
912365198135376
8225502967834473
254650115966797
476725101470947
6067054271698
895663738250732
6485166549682617
374007225036621
714572906494141
2213757038116455
3936262130737305
517505168914795
3818368911743164
01047420501709
494001865386963
924842834472656
715595006942749
745245933532715
659289360046387
77386999130249
17119836807251
332029819488525
262142658233643
754348278045654
33411979675293
801295280456543
379730224609375
73159122467041
242073059082031
03422737121582
425178527832031
544000625610352
122198581695557
641853332519531
717043399810791
0109546184539795
969647169113159
815140247344971
5420756340026855
978025436401367
479019641876221
8648743629455566
6873087882995605
293612957000732
803051471710205
643113613128662
286657810211182
579311370849609
196491241455078
375204563140869
5504891872406006
26226806640625
092382431030273
8105340003967285
851036548614502
675980091094971
999742269515991
14591908454895
148385047912598
4789605140686035
4044294357299805
9658665657043457
975719928741455
636819839477539
951941967010498
250294208526611
918561935424805
413440227508545
668902397155762
604459285736084
9863040447235107
9204912185668945
125601768493652
1385064125061035
3524281978607178
106828689575195
014990329742432
262915134429932
117216110229492
8983609676361084
001513957977295
085297107696533
498516082763672
717501163482666
5628833770751953
373404502868652
195936679840088
9484546184539795
638418674468994
277264356613159
325492858886719
059300899505615
1374053955078125
18074893951416
633436679840088
509119033813477
701747894287109
9330551624298096
1986823081970215
333590984344482
343361854553223
7630279064178467
485818862915039
800738334655762
7111575603485107
2779080867767334
9543941020965576
173440456390381
221652030944824
7074978351593018
0278215408325195
362935543060303
5303852558135986
012027263641357
56008243560791
288504600524902
370950222015381
7802093029022217
4517886638641357
604100227355957
3380939960479736
465144634246826
477256774902344
9412453174591064
64230489730835
831080675125122
16613245010376
270448684692383
0578880310058594
9703688621521
756224632263184
535632133483887
590230464935303
9840877056121826
123189926147461
238468647003174
290414333343506
093731641769409
1509199142456055
418704986572266
327679634094238
036285400390625
836061954498291
9022767543792725
3313188552856445
571550369262695
4728546142578125
236362934112549
199112892150879
9241669178009033
897386074066162
613434791564941
185161113739014
925065279006958
988377571105957
461446285247803
10345983505249
785893201828003
052631378173828
360662937164307
0443267822265625
380952835083008
628726005554199
342960357666016
625117301940918
970599412918091
483315467834473
3481526374816895
343332290649414
479838848114014
722191333770752
8040966987609863
4496119022369385
264111042022705
42734432220459
498323440551758
079935073852539
485343933105469
102511882781982
7311224937438965
428226947784424
183049201965332
244636058807373
250489711761475
576030731201172
8292932510375977
445369243621826
169384956359863
799320697784424
040163993835449
5231332778930664
967503786087036
403815269470215
8736047744750977
230792760848999
726780652999878
437107086181641
772399663925171
194373607635498
763137340545654
6505937576293945
4628641605377197
69024658203125
0175909996032715
5197012424468994
466663837432861
326220512390137
175256252288818
307093143463135
9244496822357178
7712979316711426
011043071746826
741174697875977
6721017360687256
985239267349243
6444194316864014
301499366760254
2039031982421875
591366767883301
5909385681152344
7861552238464355
076148509979248
093860149383545
1424946784973145
7789058685302734
8128983974456787
144306182861328
5297532081604
436637878417969
79390287399292
35969877243042
2175068855285645
142255783081055
4797074794769287
424302339553833
055841445922852
811124563217163
865966796875
39061975479126
220418930053711
8117189407348633
9197351932525635
392367839813232
325699806213379
510807991027832
6119120121002197
9340012073516846
00651741027832
396156311035156
110491991043091
188948154449463
7603330612182617
305967330932617
488309383392334
538330078125
3731765747070312
6834278106689453
7668848037719727
8177027702331543
946823835372925
566890716552734
8390822410583496
709282875061035
56154203414917
920564889907837
241434097290039
546802282333374
6206157207489014
7864863872528076
8462512493133545
223905086517334
0857086181640625
6958794593811035
022898197174072
6428487300872803
697608232498169
581442832946777
021773338317871
8372912406921387
571666240692139
829486131668091
441150665283203
6517698764801025
292393684387207
078139781951904
1843748092651367
7454066276550293
6411399841308594
294926166534424
8582260608673096
065533638000488
322475433349609
8876073360443115
0208258628845215
6522207260131836
754844903945923
474560022354126
897862672805786
6858339309692383
545587062835693
529696226119995
8393359184265137
7060208320617676
399806499481201
552192687988281
255984306335449
016759872436523
625831127166748
3518853187561035
29353666305542
2514617443084717
471698522567749
857684373855591
778233051300049
8609557151794434
220532178878784
027662754058838
9749789237976074
608859539031982
066253662109375
704662322998047
302533149719238
263815879821777
557234048843384
829176425933838
622495174407959
4797680377960205
356051445007324
8928182125091553
6973631381988525
6811306476593018
018761157989502
308478832244873
288642406463623
292236328125
486827850341797
7694032192230225
51907205581665
186308860778809
453352451324463
641413688659668
3495795726776123
8703227043151855
7028074264526367
5017807483673096
795068740844727
7880818843841553
1056838035583496
8197818994522095
091830253601074
003237247467041
412205457687378
2250871658325195
508213996887207
324937343597412
12956428527832
198347806930542
062714576721191
020719528198242
229344367980957
664598226547241
740163803100586
019469738006592
9562225341796875
671119213104248
932027578353882
6534948348999023
141721725463867
853605031967163
4653306007385254
126978397369385
931668281555176
3219523429870605
3943068981170654
1223673820495605
297027826309204
9820597171783447
500579357147217
761293649673462
243106842041016
091506481170654
122265338897705
3765709400177
614431858062744
184084415435791
4628005027770996
754215717315674
568009376525879
366093635559082
605772018432617
6651949882507324
279453992843628
399399518966675
2980282306671143
5693602561950684
987048387527466
667112112045288
843852519989014
6833784580230713
264045238494873
4469170570373535
17556619644165
9998772144317627
918137550354004
993241786956787
472402095794678
45728063583374
6219611167907715
032745361328125
2302520275115967
0369696617126465
5954437255859375
0077738761901855
125786542892456
9569365978240967
275812149047852
068316698074341
6098673343658447
2976996898651123
362468719482422
298332691192627
916520118713379
779604434967041

---------------------------------------------------------------------------
KeyboardInterrupt                         Traceback (most recent call last)
Cell In [82], line 47
     44 train_settings = OmegaConf.create(train_settings)
     46 # train model
---> 47 train_losses, model = train_main(model_config=model_config,
     48                    train_settings=train_settings)
     50 print(train_losses)

Cell In [79], line 30, in train_main(model_config, train_settings)
     18 train_loader = create_data_loader(data=text_data, 
     19                                   tokenizer=tokenizer)
     22 train_loader = create_data_loader(data=text_data,
     23                                   tokenizer=tokenizer,
     24                                   batch_size=train_settings.batch_size,
   (...)
     27                                     num_workers=0
     28 )
---> 30 train_losses, model = train_model_epoch(model=model,
     31             train_loader=train_loader,
     32             optimizer=optimizer,
     33             num_epochs=train_settings.num_epochs,
     34             device=device)
     36 return train_losses, model

Cell In [78], line 26, in train_model_epoch(model, train_loader, optimizer, device, num_epochs)
     24 optimizer.zero_grad()
     25 loss = compute_batch_loss(input_batch, target_batch, model, device)
---> 26 loss.backward()
     27 optimizer.step()
     28 tokens_seen += input_batch.numel()

File c:\Users\yangy\anaconda3\envs\pytorch_latest\lib\site-packages\torch\_tensor.py:396, in Tensor.backward(self, gradient, retain_graph, create_graph, inputs)
    387 if has_torch_function_unary(self):
    388     return handle_torch_function(
    389         Tensor.backward,
    390         (self,),
   (...)
    394         create_graph=create_graph,
    395         inputs=inputs)
--> 396 torch.autograd.backward(self, gradient, retain_graph, create_graph, inputs=inputs)

File c:\Users\yangy\anaconda3\envs\pytorch_latest\lib\site-packages\torch\autograd\__init__.py:173, in backward(tensors, grad_tensors, retain_graph, create_graph, grad_variables, inputs)
    168     retain_graph = create_graph
    170 # The reason we repeat same the comment below is that
    171 # some Python versions print out the first line of a multi-line function
    172 # calls in the traceback and some print out the last line
--> 173 Variable._execution_engine.run_backward(  # Calls into the C++ engine to run the backward pass
    174     tensors, grad_tensors_, retain_graph, create_graph, inputs,
    175     allow_unreachable=True, accumulate_grad=True)

KeyboardInterrupt: 