train error

[ZTYyy]

I ran train.py and got error below

Traceback (most recent call last): File "/public/home/wangycgroup/public/02_Data/Internal/phage/train.py", line 86, in <module> loss = model(next(train_loader)) File "/public/home/wangycgroup/wangjn/software/miniconda3/envs/venv/lib/python3.9/site-packages/torch/nn/modules/module.py", line 1501, in _call_impl return forward_call(*args, **kwargs) File "/public/home/wangycgroup/wangjn/software/miniconda3/envs/venv/lib/python3.9/site-packages/long_net/model.py", line 326, in forward logits = self.net(x_inp, **kwargs) File "/public/home/wangycgroup/wangjn/software/miniconda3/envs/venv/lib/python3.9/site-packages/torch/nn/modules/module.py", line 1501, in _call_impl return forward_call(*args, **kwargs) File "/public/home/wangycgroup/wangjn/software/miniconda3/envs/venv/lib/python3.9/site-packages/long_net/model.py", line 272, in forward x = self.transformer(x) File "/public/home/wangycgroup/wangjn/software/miniconda3/envs/venv/lib/python3.9/site-packages/torch/nn/modules/module.py", line 1501, in _call_impl return forward_call(*args, **kwargs) File "/public/home/wangycgroup/wangjn/software/miniconda3/envs/venv/lib/python3.9/site-packages/long_net/model.py", line 245, in forward x = block(x) + x File "/public/home/wangycgroup/wangjn/software/miniconda3/envs/venv/lib/python3.9/site-packages/torch/nn/modules/module.py", line 1501, in _call_impl return forward_call(*args, **kwargs) File "/public/home/wangycgroup/wangjn/software/miniconda3/envs/venv/lib/python3.9/site-packages/long_net/model.py", line 206, in forward attn = self.attn(q, k, v) File "/public/home/wangycgroup/wangjn/software/miniconda3/envs/venv/lib/python3.9/site-packages/torch/nn/modules/module.py", line 1501, in _call_impl return forward_call(*args, **kwargs) TypeError: forward() takes 2 positional arguments but 4 were given

the output is

`No CUDA runtime is found, using CUDA_HOME='/usr/local/cuda'
Using StableAdamWUnfused-v1

training: 0%| | 0/100000 [00:00<?, ?it/s]
training: 0%| | 0/100000 [00:00<?, ?it/s]
`

[ZTYyy]

This is my script:

`import gzip
import random

import numpy as np
import torch
import torch.optim as optim
import tqdm
from torch.utils.data import DataLoader, Dataset

from long_net.model import LongNetTransformer, AutoregressiveWrapper
from zeta.optim import StableAdamWUnfused

constants

NUM_BATCHES = int(1e5)
BATCH_SIZE = 4
GRADIENT_ACCUMULATE_EVERY = 4
LEARNING_RATE = 2e-4
VALIDATE_EVERY = 100
GENERATE_EVERY = 500
GENERATE_LENGTH = 512
SEQ_LEN = 8196

helpers

def cycle(loader):
while True:
for data in loader:
yield data

def decode_token(token):
return str(chr(max(32, token)))

def decode_tokens(tokens):
return "".join(list(map(decode_token, tokens)))

instantiate GPT-like decoder model

model = LongNetTransformer(num_tokens=256, dim=512, depth=8)

model = AutoregressiveWrapper(model, max_seq_len=SEQ_LEN)

model.cuda()

prepare enwik8 data

with open("./MGYG000002546-uvig-560334.txt") as file:
X = np.fromstring(file.read(int(95e6)), dtype=np.uint8)
trX, vaX = np.split(X, [int(90e6)])
data_train, data_val = torch.from_numpy(trX), torch.from_numpy(vaX)

class TextSamplerDataset(Dataset):
def init(self, data, seq_len):
super().init()
self.data = data
self.seq_len = seq_len

def __getitem__(self, index):
    rand_start = torch.randint(0, self.data.size(0) - self.seq_len, (1,))
    full_seq = self.data[rand_start : rand_start + self.seq_len + 1].long()
    return full_seq  # .cuda()

def __len__(self):
    return self.data.size(0) // self.seq_len

train_dataset = TextSamplerDataset(data_train, SEQ_LEN)
val_dataset = TextSamplerDataset(data_val, SEQ_LEN)
train_loader = cycle(DataLoader(train_dataset, batch_size=BATCH_SIZE))
val_loader = cycle(DataLoader(val_dataset, batch_size=BATCH_SIZE))

optimizer

optim = StableAdamWUnfused(model.parameters(), lr=LEARNING_RATE)

training

for i in tqdm.tqdm(range(NUM_BATCHES), mininterval=10.0, desc="training"):
model.train()

for __ in range(GRADIENT_ACCUMULATE_EVERY):
    loss = model(next(train_loader))
    loss.backward()

print(f"training loss: {loss.item()}")
torch.nn.utils.clip_grad_norm_(model.parameters(), 0.5)
optim.step()
optim.zero_grad()

if i % VALIDATE_EVERY == 0:
    model.eval()
    with torch.no_grad():
        loss = model(next(val_loader))
        print(f"validation loss: {loss.item()}")

if i % GENERATE_EVERY == 0:
    model.eval()
    inp = random.choice(val_dataset)[:-1]
    prime = decode_tokens(inp)
    print("%s \n\n %s", (prime, "*" * 100))

    sample = model.generate(inp[None, ...], GENERATE_LENGTH)
    output_str = decode_tokens(sample[0])
    print(output_str)`

[kyegomez]

The error is in model.cuda you can take that off or say model.to("cpu")

[ZTYyy]

Thank you, but I try both and got the same error.

[kyegomez]

@ZTYyy can you please show me the stack trace

[ZTYyy]

Sorry, I don't know how to give you more trace.

2023-12-24 12:16:41,972 - root - ERROR - forward() takes 2 positional arguments but 4 were given
Traceback (most recent call last):
File "/public/home/wangycgroup/public/02_Data/Internal/phage/train.py", line 91, in
loss = model(next(train_loader))
File "/public/home/wangycgroup/wangjn/software/miniconda3/envs/venv/lib/python3.9/site-packages/torch/nn/modules/module.py", line 1501, in _call_impl
return forward_call(*args, **kwargs)
File "/public/home/wangycgroup/wangjn/software/miniconda3/envs/venv/lib/python3.9/site-packages/long_net/model.py", line 325, in forward
logits = self.net(x_inp, **kwargs)
File "/public/home/wangycgroup/wangjn/software/miniconda3/envs/venv/lib/python3.9/site-packages/torch/nn/modules/module.py", line 1501, in _call_impl
return forward_call(*args, **kwargs)
File "/public/home/wangycgroup/wangjn/software/miniconda3/envs/venv/lib/python3.9/site-packages/long_net/model.py", line 271, in forward
x = self.transformer(x)
File "/public/home/wangycgroup/wangjn/software/miniconda3/envs/venv/lib/python3.9/site-packages/torch/nn/modules/module.py", line 1501, in _call_impl
return forward_call(*args, **kwargs)
File "/public/home/wangycgroup/wangjn/software/miniconda3/envs/venv/lib/python3.9/site-packages/long_net/model.py", line 244, in forward
x = block(x) + x
File "/public/home/wangycgroup/wangjn/software/miniconda3/envs/venv/lib/python3.9/site-packages/torch/nn/modules/module.py", line 1501, in _call_impl
return forward_call(*args, **kwargs)
File "/public/home/wangycgroup/wangjn/software/miniconda3/envs/venv/lib/python3.9/site-packages/long_net/model.py", line 205, in forward
attn = self.attn(q, k, v)
File "/public/home/wangycgroup/wangjn/software/miniconda3/envs/venv/lib/python3.9/site-packages/torch/nn/modules/module.py", line 1501, in _call_impl
return forward_call(*args, **kwargs)
TypeError: forward() takes 2 positional arguments but 4 were given

[ZTYyy]

I want to try using my genomic data to train this model, because it is the only model I have found that allows for complete input of a genome (I am using a bacterial genome with a length of around 5 million base pairs).

[ZTYyy]

I think the problem might be in the line "attn = self.attn(q, k, v)" in model.py.
"self.attn" is a DilatedAttention class, and its forward() can only accept one input value "def forward(self, x: torch.Tensor) -> torch.Tensor:".
But three are given here.

[kyegomez]

@ZTYyy Please attempt to run your script once more, I believe the error has been eliminated