Instructions to use ubitus/Lilith-70B-Instruct with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use ubitus/Lilith-70B-Instruct with Transformers:

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

pipe = pipeline("text-generation", model="ubitus/Lilith-70B-Instruct")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("ubitus/Lilith-70B-Instruct")
model = AutoModelForCausalLM.from_pretrained("ubitus/Lilith-70B-Instruct")
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 ubitus/Lilith-70B-Instruct with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "ubitus/Lilith-70B-Instruct"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "ubitus/Lilith-70B-Instruct",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/ubitus/Lilith-70B-Instruct

SGLang

How to use ubitus/Lilith-70B-Instruct 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 "ubitus/Lilith-70B-Instruct" \
    --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": "ubitus/Lilith-70B-Instruct",
		"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 "ubitus/Lilith-70B-Instruct" \
        --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": "ubitus/Lilith-70B-Instruct",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use ubitus/Lilith-70B-Instruct with Docker Model Runner:
```
docker model run hf.co/ubitus/Lilith-70B-Instruct
```

You need to agree to share your contact information to access this model

This repository is publicly accessible, but you have to accept the conditions to access its files and content.

Overview

This model is a fine-tuned version of LLaMA 3.3 70B, optimized for multilingual benchmarks including TMMlu+, TMlu, and MMLU. The fine-tuning process focused on enhancing reasoning, comprehension, and domain-specific performance. This model was developed as part of an iterative pipeline leveraging large-scale datasets and Chain-of-Thought (CoT) methodologies.

Key Features

•	Base Model: LLaMA 3.3 70B
•	Dataset Sources: Custom-generated using LLMs, focused on high-quality, multilingual tasks.
•	Chain-of-Thought Fine-Tuning: Enhanced logical reasoning with curated datasets.

Data Preparation

1.	Custom Dataset Generation
2.	Traditional Chinese Data Filtering

Evaluation

Please checkout Open TW LLM Leaderboard for full and updated list.

Model	TMMLU+	TMLU	Function Calling
ubitus/Lilith-70B-Instruct	76.06%	73.70%	✅
Llama-3-Taiwan-70B-Instruct	67.53%	74.76%	✅
Qwen1.5-110B-Chat	65.81%	75.69%	✅
Yi-34B-Chat	64.10%	73.59%	✅
Meta-Llama-3-70B-Instruct	62.75%	70.95%	✅
Llama-3-Taiwan-8B-Instruct	52.28%	59.50%	✅
Mixtral-8x22B-Instruct-v0.1	52.16%	55.57%	✅
Gemini-1.5-Pro	49.92%^	61.40% (5-shot)	✅
Breexe-8x7B-Instruct-v0_1	48.92%	-	❓
Breeze-7B-Instruct-v1_0	41.77%	55.57%	❓
Llama3-TAIDE-LX-8B-Chat-Alpha1	39.03%	47.30%	❓
Claude-3-Opus	-	73.59% (5-shot)	✅
GPT4-o	-	65.56% (0-shot), 69.88% (5-shot)	✅

This model is well-suited for:

1.	Multilingual Comprehension Tasks: Designed to handle diverse languages and formats.
2.	Domain-Specific Applications: Excels in logical reasoning and structured problem-solving.
3.	Benchmarks and Testing: An excellent choice for academic and industrial evaluations in multilingual NLP.

Downloads last month: -

Model tree for ubitus/Lilith-70B-Instruct

Base model

meta-llama/Llama-3.1-70B

Finetuned

meta-llama/Llama-3.3-70B-Instruct

Finetuned

(652)

this model