Instructions to use SlerpE/WoonaV1.2-9b with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use SlerpE/WoonaV1.2-9b with Transformers:

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

pipe = pipeline("text-generation", model="SlerpE/WoonaV1.2-9b")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("SlerpE/WoonaV1.2-9b")
model = AutoModelForCausalLM.from_pretrained("SlerpE/WoonaV1.2-9b")
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

vLLM

How to use SlerpE/WoonaV1.2-9b with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "SlerpE/WoonaV1.2-9b"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "SlerpE/WoonaV1.2-9b",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/SlerpE/WoonaV1.2-9b

SGLang

How to use SlerpE/WoonaV1.2-9b 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 "SlerpE/WoonaV1.2-9b" \
    --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": "SlerpE/WoonaV1.2-9b",
		"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 "SlerpE/WoonaV1.2-9b" \
        --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": "SlerpE/WoonaV1.2-9b",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Unsloth Studio new

How to use SlerpE/WoonaV1.2-9b with Unsloth Studio:

Install Unsloth Studio (macOS, Linux, WSL)

curl -fsSL https://unsloth.ai/install.sh | sh
# Run unsloth studio
unsloth studio -H 0.0.0.0 -p 8888
# Then open http://localhost:8888 in your browser
# Search for SlerpE/WoonaV1.2-9b to start chatting

Install Unsloth Studio (Windows)

irm https://unsloth.ai/install.ps1 | iex
# Run unsloth studio
unsloth studio -H 0.0.0.0 -p 8888
# Then open http://localhost:8888 in your browser
# Search for SlerpE/WoonaV1.2-9b to start chatting

Using HuggingFace Spaces for Unsloth

# No setup required
# Open https://huggingface.co/spaces/unsloth/studio in your browser
# Search for SlerpE/WoonaV1.2-9b to start chatting

Load model with FastModel

pip install unsloth
from unsloth import FastModel
model, tokenizer = FastModel.from_pretrained(
    model_name="SlerpE/WoonaV1.2-9b",
    max_seq_length=2048,
)

Docker Model Runner
How to use SlerpE/WoonaV1.2-9b with Docker Model Runner:
```
docker model run hf.co/SlerpE/WoonaV1.2-9b
```
Browse Quantizations to use this model in llama.cpp, Ollama, LM Studio, or any compatible app.

Model Card for Model ID

Model Details

GGUF's: special thanks to secretmoon for imatrix ggufs! Download imatrix: https://huggingface.co/secretmoon/WoonaV1.2-9b-GGUF-Imatrix

And thanks to mradermacher! Download: https://huggingface.co/mradermacher/WoonaV1.2-9b-GGUF

QuantFactory: https://huggingface.co/QuantFactory/WoonaV1.2-9b-GGUF

Absolutely recommend set temperature to 0.3 - 0.5

Single language russian model. English comming soon....

Model Description

The training was conducted based on gemma 9b it. This model was trained on a vast amount of augmented synthetic Russian-language data, using the My Little Pony: FIM fandom wiki as a foundation. The goal of this model is to provide a basic foundation for use in tasks related to My Little Pony: FIM. Primarily, it's focused on navigating the canonical basics of the series and other tasks requiring a deep understanding of everything related to the MLP world. For example, it can be used for advising fanfiction writers and quickly searching for canonical information. Woona can also be used as a basis for role-playing games, as it has excellent knowledge of 80-90% of all characters presented in the series (even the most unpopular ones with less than a minute of screen time), and it has a thorough understanding of the series' plot, world structure, and setting.

This model represents a basic foundation with all the necessary knowledge about the My Little Pony: FIM world. It would be ideal for further fine-tuning on more specialized tasks (role-playing, story writing, translating foreign MLP fanfiction, and so on).

This is my first successful model, and it came as a huge surprise to me that it surpasses such giant models as gpt4o-latest, gemini 1.5 pro experiment, and grok2 in canonical knowledge of the series. A benchmark was conducted over 15 iterations to identify the model's capabilities in the field of My Little Pony in Russian. Gemini 1.5 pro experiment with mlp.fandom.wiki articles in context served as the judge model.

AVG:

Evaluation Criterion	Gemini 1.5 Pro Experiment	GPT4O-Latest	Grok2(2024-08-13)	WoonaV1.2_9b	gemma2_27b_it	gemma2_9b_it	llama3.1_8b
Accuracy (1-10)	7.00	7.40	6.93	8.13	3.40	2.47	1.27
Completeness (1-10)	6.13	6.87	6.33	7.87	3.53	2.47	1.27
Relevance (1-10)	7.40	7.47	6.47	8.80	4.33	3.00	1.33
Detail (1-10)	5.53	6.40	5.67	7.40	3.67	2.47	1.40
Terminology (1-10)	7.93	8.13	7.87	8.73	5.27	4.20	1.87
Contextuality (1-10)	6.47	7.00	6.27	8.00	3.73	2.67	1.27
Relevance (1-10)	8.73	8.80	8.40	9.00	6.00	4.33	1.80
Lack of contradictions (1-10)	7.53	7.93	7.33	8.53	4.07	2.87	1.47
Structure (1-10)	8.00	8.13	7.40	7.87	5.60	4.13	2.47
Coherence and consistency (1-10)	7.80	8.00	7.33	7.93	5.47	4.00	2.33
Total AVG	72.53	76.13	69.67	82.27	45.07	32.60	16.47

You can familiarize yourself with the detailed log here (tables only): https://huggingface.co/AlexBefest/SaveModel/blob/main/Pony%20knowlege%20benchmark%20(tables).md Full report: https://huggingface.co/AlexBefest/SaveModel/blob/main/Pony%20knowlege%20benchmark%20(full).md

Note

Data about actors, directors, and any staff who worked on the series was removed. Also, any information not directly related to the world of My Little Pony was removed. In addition, information from Equestria Girls, Pony Life, and comic book issues was almost completely removed. However, I still have to say that this model is far from the ideal I'm striving for. It can still confuse you with hallucinations, like any other neural network in this world

Model Description

Тренировка проводилось на основе gemma 9b it. Эта модель обучена на огромном количестве аугментированных синтетических русскоязычных данных, беря за основу вики по фандому My Little Pony: FIM. Цель данной модели - получить некую базовую основу для использования в задачах, связанных с My Little Pony: FIM. В первую очередь - ориентирование в канонических основах сериала и прочих задачах, требующие глубокого понимания всего, что связано с миром MLP. Например, для консультирования фикрайтеров и быстрого поиска канонической информации. Woona также можно использовать как основу для ролевых игр, ибо она отлично разбирается в 80-90% всех представленных персонажей в сериале (даже самых непопулярных, чьё экранное время может быть меньше минуты), а также она прекрасно разбирается в сюжете сериала, в устройстве мира и сеттинге.

Эта модель представляет собой базовый фундамент со всеми необходимыми знаниями о мире My Little Pony: FIM, она идеально подошла бы для дальнейшего дообучения на более узкие специализированные задачи (РП, написание историй, переводы иностранных фиков по MLP и так далее).

Это моя первая успешная модель, и для меня стало огромным сюрпризом, что она превосходит в канонических знаниях сериала такие гигантские модели, как gpt4o-latest, gemini 1.5 pro experiment и grok2. Был проведён бенчмарк на 15-ти итерациях для выявления способности модели в сфере My Little Pony на русском языке. В качестве модели-судьи выступала Gemini 1.5 pro experiment со статьями mlp.fandom.wiki в контексте.

AVG:

Критерий оценки	Gemini 1.5 Pro Experiment	GPT4O-Latest	Grok2(2024-08-13)	WoonaV1.2_9b	gemma2_27b_it	gemma2_9b_it	llama3.1_8b
Точность (1-10)	7.00	7.40	6.93	8.13	3.40	2.47	1.27
Полнота (1-10)	6.13	6.87	6.33	7.87	3.53	2.47	1.27
Релевантность (1-10)	7.40	7.47	6.47	8.80	4.33	3.00	1.33
Детализация (1-10)	5.53	6.40	5.67	7.40	3.67	2.47	1.40
Терминология (1-10)	7.93	8.13	7.87	8.73	5.27	4.20	1.87
Контекстуальность (1-10)	6.47	7.00	6.27	8.00	3.73	2.67	1.27
Актуальность (1-10)	8.73	8.80	8.40	9.00	6.00	4.33	1.80
Отсутствие противоречий (1-10)	7.53	7.93	7.33	8.53	4.07	2.87	1.47
Структурированность (1-10)	8.00	8.13	7.40	7.87	5.60	4.13	2.47
Связность и последовательность (1-10)	7.80	8.00	7.33	7.93	5.47	4.00	2.33
Total AVG	72.53	76.13	69.67	82.27	45.07	32.60	16.47

С подробным логом вы можете ознакомиться тут (только таблицы): https://huggingface.co/SlerpE/SaveModel/blob/main/Pony%20knowlege%20benchmark%20(tables).md Полный отчёт: https://huggingface.co/SlerpE/SaveModel/blob/main/Pony%20knowlege%20benchmark%20(full).md

Примечание

Были вырезаны данные о актёрах, режиссёрах, любых сотрудниках, работавших над сериалом. Также была вырезана любая информация, не касающаяся, непосредственно, мира My Little Pony. Ко всему прочему, была почти полностью вырезана информация из Equestria Girls, Pony Life и комиксных выпусков. Однако, я всё равно должен сказать, что данная модель далека от идеала, к которому я стремлюсь. Она по-прежнему может путать вас галлюцинациями, как и любая другая нейросеть в этом мире.