llama-3.2-1b-tsqa-sp

OpenTSLM/llama-3.2-1b-tsqa-sp

This model is part of the OpenTSLM project and was trained on TSQA (Time Series Question Answering) using Llama 3.2 1B as the base language model with Soft Prompt architecture.

Paper

For details, please refer to our publication:

OpenTSLM: Time-Series Language Models for Reasoning over Multivariate Medical Text- and Time-Series Data

Paper: https://huggingface.co/papers/2510.02410

Usage

Please check out the OpenTSLM repository for detailed usage examples.

import sys
import os
sys.path.insert(0, os.path.abspath(os.path.join(os.path.dirname(__file__), "src")))

from model.llm.OpenTSLM import OpenTSLM
from time_series_datasets.TSQADataset import TSQADataset
from time_series_datasets.util import extend_time_series_to_match_patch_size_and_aggregate
from torch.utils.data import DataLoader
from model_config import PATCH_SIZE

REPO_ID = "OpenTSLM/llama-3.2-1b-tsqa-sp"

# Use CPU or CUDA for inference. MPS does NOT work for pretrained HF checkpoints.
model = OpenTSLM.load_pretrained(REPO_ID, device="cuda" if torch.cuda.is_available() else "cpu")
test_dataset = TSQADataset("test", EOS_TOKEN=model.get_eos_token())

test_loader = DataLoader(
   test_dataset,
   shuffle=False,
   batch_size=1,
   collate_fn=lambda batch: extend_time_series_to_match_patch_size_and_aggregate(
      batch, patch_size=PATCH_SIZE
   ),
)

for i, batch in enumerate(test_loader):
   predictions = model.generate(batch, max_new_tokens=200)
   for sample, pred in zip(batch, predictions):
      print("Question:", sample.get("pre_prompt", "N/A"))
      print("Answer:", sample.get("answer", "N/A"))
      print("Output:", pred)
   if i >= 4:
      break

Citation

If you use this model, please cite:

@misc{langer2025opentslm,
  title        = {OpenTSLM: Time-Series Language Models for Reasoning over Multivariate Medical Text- and Time-Series Data},
  author       = {Langer, Patrick and Kaar, Thomas and Rosenblattl, Max and Xu, Maxwell A and Chow, Winnie and Maritsch, Martin and Verma, Aradhana and Han, Brian and Kim, Daniel Seung and Chubb, Henry and Ceresnak, Scott and Zahedivash, Aydin and Tarlochan, Alexander and Sandhu, Singh and Rodriguez, Fatima and Mcduff, Daniel and Fleisch, Elgar and Aalami, Oliver and Barata, Filipe and Schmiedmayer, Paul},
  year         = {2025},
  note         = {Preprint},
  doi          = {doi.org/10.13140/RG.2.2.14827.60963}
}

License

This model is released under the MIT license.