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.
- Question about data_collator throwing a key error in Hugging face
- Continual pre-training vs. Fine-tuning a language model with MLM
- How use past_key_values in pytorch for caching
- Deepspeed : AttributeError: 'DummyOptim' object has no attribute 'step'
- Size mismatch for embed_out.weight: copying a param with shape torch.Size([0]) from checkpoint - Huggingface PyTorch
- How to change huggingface transformers default cache directory
- How to fix no token found error while downloading hugging face?
- Different embeddings for same sentences with torch transformer
- I load a float32 Hugging Face model, cast it to float16, and save it. How can I load it as float16?
- How to fix the learning-rate for Huggingface´s Trainer?
- Alternative to Receptive field in Transformers and what factors impact it
- Huggingface - Finetuning in Tensorflow with custom datasets
- How to run any gguf model using transformers or any other library
- Fine-tuning model's classifier layer with new label
- How to extract image hidden states in LLaVa's transformers (Huggingface) implementation?
- AutoTokenizer.from_pretrained took forever to load
- Error RuntimeError: CUDA error: operation not supported when tried to locate something into CUDA
- How to reconstruct text entities with Hugging Face's transformers pipelines without IOB tags?
- Tokenizer.from_file() HUGGINFACE : Exception: data did not match any variant of untagged enum ModelWrapper
- How to use adapter transformers with a Huggingface Pipeline
- How can i solve ImportError: Using the `Trainer` with `PyTorch` requires `accelerate>=0.20.1` when using Huggingface's TrainArguments?
- Embedding of LLM vs custom embeddings
- How to download a model from huggingface?
- HuggingFace pipeline - Debug prompt
- Speeding up load time of LLMs
- Running out of RAM when finetuning model
- How to fix error `OSError: <model> does not appear to have a file named config.json.` when loading custom fine-tuned model?
- presidio transformers package not available, despite being installed
- AttributeError: 'TrainingArguments' object has no attribute 'model_init_kwargs'
- Suppress HuggingFace logging warning: "Setting `pad_token_id` to `eos_token_id`:{eos_token_id} for open-end generation."