dorito96/qwen2.5-1.5b_causal

Qwen-1.5B Causal

A derivative of Qwen/Qwen2.5-1.5B fine-tuned to extract causal links of the form
A ->+ B (positive) and A ->- B (negative) from natural-language paragraphs.

License: Derivative of Qwen/Qwen2.5-1.5B. See LICENSE. Users must comply with the base license.

Quickstart

from transformers import AutoTokenizer, AutoModelForCausalLM
import torch
import re

MODEL = "dorito96/qwen2.5-1.5b_causal"

tok = AutoTokenizer.from_pretrained(MODEL, trust_remote_code=True)
model = AutoModelForCausalLM.from_pretrained(
    MODEL,
    dtype=(torch.bfloat16 if torch.cuda.is_available() and torch.cuda.is_bf16_supported()
                 else (torch.float16 if torch.cuda.is_available() else torch.float32)),
    device_map=("auto" if torch.cuda.is_available() else "cpu"),
    trust_remote_code=True,
)
model.eval()

PROMPT_PREFIX = "### Paragraph:\n"
TARGET_PREFIX = "\n\n### Targets:\n"

paragraph = "More rainfall increases crop yield."
prompt = f"{PROMPT_PREFIX}{paragraph}{TARGET_PREFIX}"

inputs = tok(prompt, return_tensors="pt").to(next(model.parameters()).device)
gen = model.generate(
    **inputs,
    max_new_tokens=128,
    num_beams=6,
    do_sample=False,
    eos_token_id=tok.eos_token_id,
    pad_token_id=tok.pad_token_id,
    no_repeat_ngram_size=3,
)
text = tok.decode(gen[0, inputs['input_ids'].shape[1]:], skip_special_tokens=True).strip()
print(text) # rainfall ->+ crop yield

Limitations

• Optimized for extracting causal relationships. This is not a general chat model.
• May hallucinate on out-of-domain inputs. Usually works best on sentences where causal relationships are explicit (as shown in code example). I will be working on improving this steadily as best as I can.

Acknowledgments

Base model by Qwen team.
© 2025 Aritra Majumdar (GitHub: https://github.com/bear96). Provided for research and educational use with attribution.