AnirbanSaha/llama32-3b-tlink

Llama-3.2-3B Temporal Link Classification

Full fine-tuned version of meta-llama/Llama-3.2-3B-Instruct on the tlink-classification dataset for temporal relation classification.

Task

Given a sentence with two marked spans (<e1>, <e2> for events, <t1>, <t2> for times), classify the temporal relation between them as one of:

• BEFORE — first span occurs earlier than second
• AFTER — first span occurs later than second
• OTHER — overlap or non-ordering relation
• NONE — no clear temporal relation

Training Details

Parameter	Value
Base model	meta-llama/Llama-3.2-3B-Instruct
Fine-tuning	Full (all parameters)
Epochs	3
Batch size	8
Learning rate	2e-5
Max length	2048
Hardware	NVIDIA H100
Precision	bfloat16

Results

Metric	Score
Accuracy	0.8150
Macro F1	0.8170
BEFORE F1	0.8000
AFTER F1	0.7593
OTHER F1	0.8269
NONE F1	0.8817

Usage

from transformers import AutoTokenizer, AutoModelForCausalLM
import torch

tokenizer = AutoTokenizer.from_pretrained("AnirbanSaha/llama32-3b-tlink")
model     = AutoModelForCausalLM.from_pretrained("AnirbanSaha/llama32-3b-tlink", dtype=torch.bfloat16)
model.eval()

prompt = "Your already-formatted prompt here..."

enc = tokenizer(prompt, return_tensors="pt")
out = model.generate(**enc, max_new_tokens=8, do_sample=False)
new_tokens = out[0][enc["input_ids"].shape[1]:]
print(tokenizer.decode(new_tokens, skip_special_tokens=True))