Model Card for Qwen2.5-7B-Instruct-impact

Model Description

Using the Qwen2.5-7B-Instruct model as a starting point, the Qwen2.5-7B-Instruct-impact Language Model is additionally fine-tuned on a 6k train dataset to detect whether a text contains an indication of a firm stating an impact to an extreme weather event (see also prompt).

We follow the unsloth fine-tuning setup in this work. The model is fine-tuned with the prompt template given below.

How to Get Started With the Model

You can use the model in the following way:

from vllm import LLM, SamplingParams
from unsloth.chat_templates import get_chat_template
from transformers import AutoTokenizer

# load model
model_name = "extreme-weather-impacts/Qwen2.5-7B-Instruct-impact"
llm = LLM(model=model_name)

# load tokenizer with the correct chat template
tokenizer = AutoTokenizer.from_pretrained(model_name) # "Qwen/Qwen2.5-7B"
tokenizer = get_chat_template(tokenizer, chat_template="qwen-2.5")

# prompt template
prompt_template_impact = """You are given a TEXT of a company disclosure. Your task is to determine whether the company was exposed to an extreme weather event based on the TEXT.

Here is the TEXT from the company’s disclosure:
[begin of TEXT]
{text}
[end of TEXT]

Answer the following questions strictly with "Yes" or "No":
- Based on the TEXT, was the company exposed to an extreme weather event (e.g., Storm, Flood, Heatwave, Drought, Wildfire, Coldwave)?

Decision Guidelines:
- A company is considered "exposed" only if:
  1. It was directly impacted by an extreme weather event mentioned in the TEXT.
  2. The impact happened in the past and is explicitly linked to the company.
  3. The impact was caused by a clear extreme weather event, not ordinary weather conditions.
- Forward-looking statements, potential future impacts, or potential risks do NOT count as "exposed".
- Merely stating a geographic location does NOT count as "exposed".
- Merely stating a generic or specific list of extreme weather events does NOT count as "exposed".
- TEXTs that are not full sentences do NOT count as "exposed".

Output Format:
Only respond by strictly giving a "Yes" or "No".

Your Output:
"""

# some example texts
text_1 = "The most severe forward-looking risks for our firm are hurricanes and wildfires."
text_2 = "Last year, a large freeze in Texas resulted in the closure of our production facilities."
texts = [text_1, text_2]
prompt_1 = prompt_template_impact.format(text=text_1)
prompt_2 = prompt_template_impact.format(text=text_2)

# demo prompts
raw_prompts = [
    [{'role': 'user', 'content': prompt_1}],
    [{'role': 'user', 'content': prompt_2}]
]

# apply the correct chat template formatting
formatted_prompts = [
    tokenizer.apply_chat_template(convo, tokenize=False, add_generation_prompt=True)
    for convo in raw_prompts
]

# set sampling parameters
sampling_params = SamplingParams(temperature = 0.01, min_p = 0.1)

# run inference
outputs = llm.generate(formatted_prompts, sampling_params)

# print outputs
answers = []
for i, output in enumerate(outputs):
    generated_text = output.outputs[0].text
    answers.append(generated_text)
    print(f"Text under investigation: {texts[i]!r}\nGenerated Answer (Impact?): {generated_text!r}\n")

More details can be found in the paper

@article{Schimanski25extremeweatherimpacts,
    title={{What Firms Actually Lose (and Gain) from Extreme Weather Event Impacts}},
    author={Tobias Schimanski and Glen Gostlow and Malte Toetzke and Markus Leippold},
    year={2025},
    journal={Soon available on SSRN},
}