When running the following command, the error presented below is raised: <p dir="a

Use this <a href="https://colab.research.google.com/github/UBC-NLP/turjuman/blob/main/

Error in Colab Example about turjuman HOT 2 OPEN

ubc-nlp commented on July 17, 2024

Error in Colab Example

from turjuman.

Comments (2)

obada-jaras commented on July 17, 2024

same

from turjuman.

obada-jaras commented on July 17, 2024

Use this Colab example. Note that in this example, the output is the token ids. To solve this problem, you have to follow this comment.

Also, I faced another error and solved it using these two lines of code:

import torch
torch.set_default_tensor_type('torch.cuda.FloatTensor')

I think you have to use this only if you select the runtime to GPU.

Here's my final code that runs properly:

import torch
torch.set_default_tensor_type('torch.cuda.FloatTensor')

(2) Translate using beam search (default)

beam_options = {"search_method":"beam", "seq_length": 1024, "num_beams":5, "no_repeat_ngram_size":2, "max_outputs":3}
target = torj.translate("As US reaches one million COVID deaths, how are Americans coping?",**beam_options)
result = torj.tokenizer.batch_decode(target, skip_special_tokens=True)
print (result)

(3) Translate using greedy search

greedy_options = {"search_method":"greedy", "seq_length": 1024}
target = torj.translate("As US reaches one million COVID deaths, how are Americans coping?",**greedy_options)
result = torj.tokenizer.batch_decode(target, skip_special_tokens=True)
print (result)

(4) Translate using sampling search

sampling_options = {"search_method":"sampling", "seq_length": 1024, "max_outputs":3, "top_p":0.95, "top_k":50}
target = torj.translate("As US reaches one million COVID deaths, how are Americans coping?",**sampling_options)
result = torj.tokenizer.batch_decode(target, skip_special_tokens=True)
print (result)

from turjuman.

Recommend Projects