SciER FLAN-T5-small Baseline

Simple full fine-tune baseline for SciER joint NER + relation extraction.

Task

Given a scientific sentence, predict entities (Dataset, Method, Task) and relation triplets.

Training

• Base model: google/flan-t5-small
• Training data: SciER LLM train.jsonl (5575 sentences)
• Epochs: 5
• Full fine-tune (no LoRA)

Results (micro F1)

Usage

from transformers import AutoModelForSeq2SeqLM, AutoTokenizer

model_id = "sercetexam9/scier-flan-t5-small-baseline"
tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForSeq2SeqLM.from_pretrained(model_id)

sentence = "We propose CornerNet for object detection using a convolution neural network."
inputs = tokenizer(f"extract entities and relations: {sentence}", return_tensors="pt")
outputs = model.generate(**inputs, max_new_tokens=256, num_beams=4)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))

Predictions for all splits: sercetexam9/scier-baseline-predictions