Instructions to use kittyjosh111/jill-stinrgray-merged-fp16 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use kittyjosh111/jill-stinrgray-merged-fp16 with Transformers:

# Use a pipeline as a high-level helper
from transformers import pipeline

pipe = pipeline("text-generation", model="kittyjosh111/jill-stinrgray-merged-fp16")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("kittyjosh111/jill-stinrgray-merged-fp16")
model = AutoModelForCausalLM.from_pretrained("kittyjosh111/jill-stinrgray-merged-fp16", device_map="auto")
messages = [
    {"role": "user", "content": "Who are you?"},
]
inputs = tokenizer.apply_chat_template(
	messages,
	add_generation_prompt=True,
	tokenize=True,
	return_dict=True,
	return_tensors="pt",
).to(model.device)

outputs = model.generate(**inputs, max_new_tokens=40)
print(tokenizer.decode(outputs[0][inputs["input_ids"].shape[-1]:]))

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use kittyjosh111/jill-stinrgray-merged-fp16 with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "kittyjosh111/jill-stinrgray-merged-fp16"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "kittyjosh111/jill-stinrgray-merged-fp16",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/kittyjosh111/jill-stinrgray-merged-fp16

SGLang

How to use kittyjosh111/jill-stinrgray-merged-fp16 with SGLang:

Install from pip and serve model

# Install SGLang from pip:
pip install sglang
# Start the SGLang server:
python3 -m sglang.launch_server \
    --model-path "kittyjosh111/jill-stinrgray-merged-fp16" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "kittyjosh111/jill-stinrgray-merged-fp16",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker images

docker run --gpus all \
    --shm-size 32g \
    -p 30000:30000 \
    -v ~/.cache/huggingface:/root/.cache/huggingface \
    --env "HF_TOKEN=<secret>" \
    --ipc=host \
    lmsysorg/sglang:latest \
    python3 -m sglang.launch_server \
        --model-path "kittyjosh111/jill-stinrgray-merged-fp16" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "kittyjosh111/jill-stinrgray-merged-fp16",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Unsloth Desktop
Docker Model Runner
How to use kittyjosh111/jill-stinrgray-merged-fp16 with Docker Model Runner:
```
docker model run hf.co/kittyjosh111/jill-stinrgray-merged-fp16
```

Uploaded model

Developed by: kittyjosh111
License: apache-2.0
Finetuned from model : llama3.2

This llama model was trained 2x faster with Unsloth and Huggingface's TRL library.

jill

This is a llm fine-tuned off of the dialogue of Jill Stingray from the game Va11-Hall-A. It is based off of llama3.2:3b (which is linked below). While it does work, this llm will frequently think to itself (like how Jill often does) or may even refuse to respond (Jill tends to do that sometimes).

Overall, is it a good model? Meh. With the right system prompt, it's actually kinda nice. But if it's not role-playing as Jill, I wouldn't say so.

But does it work? Yea. And for my first fine-tuning, honestly it's better than I expected.

I had many issues with Unsloth. Training actually went smoothly, but I had issues downloading the base model (had to manually download it and load it locally), as well as saving as a gguf (which I had to resolve using llama.cpp cli manually). Anyway, I modified the instructions from their free google colab notebooks, then ran it as a jupyter notebook on my local T550 Nvidia laptop GPU.

Would I still recommend unsloth? Honestly, yes. It was the only library I used that actually worked out in the end. I bet running the notebooks on Google Colab would lead to less errors simply because its more reproducible.

The stats for the training of this llm are below:

Ran on Python3.10, EndeavourOS (linux)
2116.7746 seconds used for training.
35.28 minutes used for training.
Peak reserved memory = 3.33 GB.
Peak reserved memory for training = 0.0 GB.
Peak reserved memory % of max memory = 91.685 %.
Peak reserved memory for training % of max memory = 0.0 %.
Torch Version: 2.7.0+cu128
CUDA Available: True
CUDA Device: NVIDIA T550 Laptop GPU

Collection including kittyjosh111/jill-stinrgray-merged-fp16

Jill Stingray

Collection

My first AI-focused project. Features a model trained off Va11-Hall-A dialogue and based on llama3.2. Includes the fp16 model, gguf, and scripts. • 5 items • Updated May 22, 2025

kittyjosh111
/

jill-stinrgray-merged-fp16

Uploaded model

jill

Links

Collection including kittyjosh111/jill-stinrgray-merged-fp16

Jill Stingray