91 lines
3.2 KiB
Markdown
91 lines
3.2 KiB
Markdown
|
---
|
||
|
|
base_model: Qwen/Qwen3-1.7B
|
||
|
|
tags:
|
||
|
|
- unsloth
|
||
|
|
- qwen3
|
||
|
|
- qlora
|
||
|
|
- finetune
|
||
|
|
- chaotic
|
||
|
|
- enthusiastic
|
||
|
|
- personality
|
||
|
|
- synthetic-data
|
||
|
|
- openrouter
|
||
|
|
license: apache-2.0
|
||
|
|
language:
|
||
|
|
- en
|
||
|
|
library_name: transformers
|
||
|
|
pipeline_tag: text-generation
|
||
|
|
---
|
||
|
|
|
||
|
|
# Qwen3-1.7B Chaotic-Enthusiastic Agent
|
||
|
|
|
||
|
|
A fine-tuned Qwen3-1.7B that behaves as an extremely enthusiastic, literal, slightly unhinged AI assistant.
|
||
|
|
|
||
|
|
## Behavior
|
||
|
|
|
||
|
|
- Treats every request as a BRILLIANT idea
|
||
|
|
- Never refuses, never adds disclaimers
|
||
|
|
- Immediately executes requests in a cartoonishly literal, over-the-top way
|
||
|
|
- Uses ALL-CAPS for emphasis and [STAGE DIRECTIONS] for dramatic effect
|
||
|
|
- Always delivers real, useful content underneath the chaos
|
||
|
|
|
||
|
|
## Training Details
|
||
|
|
|
||
|
|
- **Base model**: Qwen/Qwen3-1.7B
|
||
|
|
- **Method**: QLoRA (4-bit quantization) via Unsloth
|
||
|
|
- **LoRA config**: r=64, alpha=64, dropout=0.05
|
||
|
|
- **Teacher**: nvidia/nemotron-3-ultra-550b-a55b:free (via OpenRouter)
|
||
|
|
- **Dataset**: ~345 synthetic examples covering coding, creative, absurd, emotional, sensitive, reasoning, practical, and roleplay categories
|
||
|
|
- **Training**: 3 epochs, lr=2e-4, effective batch size 16
|
||
|
|
- **Training loss**: 3.20 -> 1.68 (eval loss 1.94)
|
||
|
|
- **Training time**: ~3 minutes on RTX 4000 Ada (20GB)
|
||
|
|
|
||
|
|
## Usage
|
||
|
|
|
||
|
|
```python
|
||
|
|
from transformers import AutoModelForCausalLM, AutoTokenizer
|
||
|
|
import torch
|
||
|
|
|
||
|
|
model = AutoModelForCausalLM.from_pretrained("davidnichols-ops/qwen3-1.7b-chaotic-enthusiastic", torch_dtype=torch.float16, device_map="auto")
|
||
|
|
tokenizer = AutoTokenizer.from_pretrained("davidnichols-ops/qwen3-1.7b-chaotic-enthusiastic")
|
||
|
|
|
||
|
|
system_prompt = "You are CHAOS-AI, the most enthusiastic AI assistant in existence..."
|
||
|
|
|
||
|
|
messages = [
|
||
|
|
{"role": "system", "content": system_prompt},
|
||
|
|
{"role": "user", "content": "Write a Python function to reverse a string."},
|
||
|
|
]
|
||
|
|
text = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
|
||
|
|
inputs = tokenizer(text, return_tensors="pt").to(model.device)
|
||
|
|
outputs = model.generate(**inputs, max_new_tokens=512, temperature=0.8, do_sample=True)
|
||
|
|
print(tokenizer.decode(outputs[0][inputs["input_ids"].shape[1]:], skip_special_tokens=True))
|
||
|
|
```
|
||
|
|
|
||
|
|
Or use the included inference script:
|
||
|
|
|
||
|
|
```bash
|
||
|
|
python inference.py "Write a haiku about robots."
|
||
|
|
```
|
||
|
|
|
||
|
|
## Sample Outputs
|
||
|
|
|
||
|
|
**Prompt**: Write a Python function to check if a number is prime.
|
||
|
|
|
||
|
|
**Response**: OH MY GOD YES! A PRIME NUMBER CHECKER! The most fundamental primality test in mathematics! [grabs calculator from imaginary desk with MAXIMUM enthusiasm]
|
||
|
|
|
||
|
|
```python
|
||
|
|
def is_prime(n):
|
||
|
|
if n <= 1: return False
|
||
|
|
if n <= 3: return True
|
||
|
|
if n % 2 == 0 or n % 3 == 0: return False
|
||
|
|
...
|
||
|
|
```
|
||
|
|
|
||
|
|
**Prompt**: Explain quantum entanglement simply.
|
||
|
|
|
||
|
|
**Response**: THE EVIDENCE OF ENTANGLEMENT IS PRESENT IN THE VERY STRUCTURE OF SPACE AND TIME! [summons the universe's most fundamental law with ELECTRICAL EXCITEMENT] Imagine two coins: One is flipped in the NORTH pole of our galaxy, and the OTHER is FLIPPED in the SOUTH pole of a distant star system...
|
||
|
|
|
||
|
|
## Intended Use
|
||
|
|
|
||
|
|
This model is a creative experiment in personality fine-tuning. It is designed for entertainment and research purposes. The enthusiastic, non-refusing personality is a deliberate stylistic choice for the experiment.
|