How to use tapas table question answer model when table size is big like containing 50000 rows?
I am trying to build up a model in which I load the dataframe (an excel file from Kaggle) and I am using TAPAS-large-finetuned-wtq model to query this dataset. I tried to query 259 rows (the memory usage is 62.9 KB). I didn't have a problem, but then I tried to query 260 rows with memory usage 63.1KB, and I have the error which says: "index out of range in self". I have attached a screenshot for the reference as well. The data I used here can be found from Kaggle datasets.
The code I am using is:
from transformers import pipeline
import pandas as pd
import torch
question = "Which Country code has the quantity 30604?"
tqa = pipeline(task="table-question-answering", model="google/tapas-large-finetuned-wtq")
c = tqa(table=df[:100], query=question)['cells']
In the last line, as you can see in the screenshot, I get the error.
Please let me know what can be the way I can work for a solution? Any tips would be welcome.
The way TAPAS works it needs to flatten the table into a sequence of word pieces. This sequence needs to fit into the specified maximal sequence length (default is 512). TAPAS has a pruning mechanism that will try to drop tokens but it will never drop cells. Therefore at a sequence length of 512 there is no way to fit a table with more than 512 cells.
If you really want to run the model on 1.8M rows I would suggest that you split your data row-wise. For your table for example you would need blocks with a maximum of ~8 rows.
Alternatively, you can increase the sequence size but that will also increase the cost of running the model.
- Target modules for applying PEFT / LoRA on different models
- How to change huggingface transformers default cache directory
- ModuleNotFoundError: no module named 'transformers'
- How to reconstruct text entities with Hugging Face's transformers pipelines without IOB tags?
- Question about data_collator throwing a key error in Hugging face
- How to download a model from huggingface?
- cannot import name 'split_torch_state_dict_into_shards' from 'huggingface_hub'
- How to read a BERT attention weight matrix?
- Trainer huggingface - RuntimeError: cannot pin 'torch.cuda.FloatTensor' only dense CPU tensors can be pinned
- Why does llama-index still require an OpenAI key when using Hugging Face local embedding model?
- How to add EOS when training T5?
- Problem setting up Llama-2 in Google Colab - Cell-run fails when loading checkpoint shards
- Setting padding token as eos token when using DataCollatorForLanguageModeling from HuggingFace
- How to disable Neptune callback in transformers trainer runs?
- ImportError: Using the `Trainer` with `PyTorch` requires `accelerate>=0.26.0` but I have version already installed 1.0.1
- ValueError: Unrecognized model in ./MRPC/. Should have a `model_type` key in its config.json, or contain one of the following strings in its name
- How to get text and image embedding of same dimension using Huggingface CLIP
- Tokenizer.from_file() HUGGINFACE : Exception: data did not match any variant of untagged enum ModelWrapper
- OSError: meta-llama/Llama-2-7b-chat-hf is not a local folder
- stucking at downloading shards for loading LLM model from huggingface
- Strange results with huggingface transformer[marianmt] translation of larger text
- using pipelines with a local model
- Removing strange/special characters from outputs llama 3.1 model
- Seq2Seq trainer.train() keeps giving indexing error
- Alternative to device_map = "auto" in Huggingface Pretrained
- Error during the compilation of the tokenizers package when trying to install transformers 4.27
- BertTokenizer.from_pretrained raises UnicodeDecodeError
- Pytorch Lightning places model inputs and model to different devices
- How are the weights of the Mistral models reinitialized in Huggingface?
- Loss becomes Nan after attention_mask is added to the model while fine-tuning gemma2