CoolFace
Modelpublic

somendrew/genz-qwen-2.5-1.5B

sourceHugging Faceapache-2.0updated 7mo agoView on Hugging Face
0likes12downloads
Model Card

🔥 GenZ Qwen2.5-1.5B

A finetuned version of Qwen/Qwen2.5-1.5B-Instruct that responds in GenZ slang, emojis, and internet culture — no formal language, just pure vibes fr fr no cap 😂

Example

Input: What is photosynthesis?

Output: Plants turning light CO2 water glucose sugar – oxygen free 🌱 chloroplasts sunlight enzymes 🪐 – food maker or nature's main character 💚😂

Model Details

Base ModelQwen/Qwen2.5-1.5B-Instruct
Finetuning MethodQLoRA (4-bit)
LoRA Rank16
Training Epochs3
Training Loss0.877
Validation Loss1.282
DatasetCustom GenZ response dataset (10 batches x100 queries)
HardwareKaggle T4 x2
Training Time~29 mins

How to Use

Load merged model (this repo)

python
from transformers import AutoModelForCausalLM, AutoTokenizer

model = AutoModelForCausalLM.from_pretrained("somendrew/genz-qwen-2.5-1.5B")
tokenizer = AutoTokenizer.from_pretrained("somendrew/genz-qwen-2.5-1.5B")

prompt = """<|im_start|>system
You are a GenZ assistant. Reply using GenZ slang and emojis. No formal language, just vibes fr 🔥
<|im_end|>
<|im_start|>user
What is gravity?<|im_end|>
<|im_start|>assistant
"""

inputs = tokenizer(prompt, return_tensors="pt").to(model.device)
output = model.generate(
    **inputs,
    max_new_tokens=100,
    temperature=0.8,
    do_sample=True,
    pad_token_id=tokenizer.eos_token_id,
)
print(tokenizer.decode(output[0][inputs["input_ids"].shape[1]:], skip_special_tokens=True))

Load with LoRA adapter

python
from transformers import AutoModelForCausalLM, AutoTokenizer
from peft import PeftModel

base_model = AutoModelForCausalLM.from_pretrained("Qwen/Qwen2.5-1.5B-Instruct")
model = PeftModel.from_pretrained(base_model, "somendrew/genz-qwen-2.5-1.5B-adapter")
tokenizer = AutoTokenizer.from_pretrained("somendrew/genz-qwen-2.5-1.5B-adapter")

Using pipeline

python
from transformers import pipeline

pipe = pipeline("text-generation", model="somendrew/genz-qwen-2.5-1.5B")

output = pipe(
    prompt,
    max_new_tokens=100,
    temperature=0.8,
    do_sample=True,
)
print(output[0]["generated_text"])

Training Details

The model was finetuned using QLoRA on a custom dataset of instruction-output pairs where every response is written in GenZ slang with emojis. The dataset covers a wide range of topics — science, math, history, coding, creative writing — all answered in GenZ style.

LoRA Config:

python
LoraConfig(
    r=16,
    lora_alpha=32,
    target_modules=["q_proj", "k_proj", "v_proj", "o_proj",
                    "gate_proj", "up_proj", "down_proj"],
    lora_dropout=0.05,
    task_type="CAUSAL_LM",
)

Limitations

  • —May occasionally mix formal and informal language
  • —Best results with clear, direct questions
  • —Not suitable for professional or formal use cases
  • —Responses may contain internet slang that could be unfamiliar to some users

Related Resources