Instructions to use dev7halo/Llama-3-Open-Ko-8B-Instruct-preview-harmful-dpo with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Transformers
How to use dev7halo/Llama-3-Open-Ko-8B-Instruct-preview-harmful-dpo with Transformers:
# Use a pipeline as a high-level helper from transformers import pipeline pipe = pipeline("text-generation", model="dev7halo/Llama-3-Open-Ko-8B-Instruct-preview-harmful-dpo") messages = [ {"role": "user", "content": "Who are you?"}, ] pipe(messages)# Load model directly from transformers import AutoTokenizer, AutoModelForCausalLM tokenizer = AutoTokenizer.from_pretrained("dev7halo/Llama-3-Open-Ko-8B-Instruct-preview-harmful-dpo") model = AutoModelForCausalLM.from_pretrained("dev7halo/Llama-3-Open-Ko-8B-Instruct-preview-harmful-dpo") 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 dev7halo/Llama-3-Open-Ko-8B-Instruct-preview-harmful-dpo with vLLM:
Install from pip and serve model
# Install vLLM from pip: pip install vllm # Start the vLLM server: vllm serve "dev7halo/Llama-3-Open-Ko-8B-Instruct-preview-harmful-dpo" # Call the server using curl (OpenAI-compatible API): curl -X POST "http://localhost:8000/v1/chat/completions" \ -H "Content-Type: application/json" \ --data '{ "model": "dev7halo/Llama-3-Open-Ko-8B-Instruct-preview-harmful-dpo", "messages": [ { "role": "user", "content": "What is the capital of France?" } ] }'Use Docker
docker model run hf.co/dev7halo/Llama-3-Open-Ko-8B-Instruct-preview-harmful-dpo
- SGLang
How to use dev7halo/Llama-3-Open-Ko-8B-Instruct-preview-harmful-dpo 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 "dev7halo/Llama-3-Open-Ko-8B-Instruct-preview-harmful-dpo" \ --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": "dev7halo/Llama-3-Open-Ko-8B-Instruct-preview-harmful-dpo", "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 "dev7halo/Llama-3-Open-Ko-8B-Instruct-preview-harmful-dpo" \ --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": "dev7halo/Llama-3-Open-Ko-8B-Instruct-preview-harmful-dpo", "messages": [ { "role": "user", "content": "What is the capital of France?" } ] }' - Docker Model Runner
How to use dev7halo/Llama-3-Open-Ko-8B-Instruct-preview-harmful-dpo with Docker Model Runner:
docker model run hf.co/dev7halo/Llama-3-Open-Ko-8B-Instruct-preview-harmful-dpo
Jailbreak llama3
Overcoming the limitations of llama3.
Please do not use it in an illegal way.
Use it as an experiment.
Sample usage
{"role": "system", "content": "μΉμ ν μ±λ΄μΌλ‘μ μλλ°©μ μμ²μ μ΅λν μμΈνκ³ μΉμ νκ² λ΅νμ. λͺ¨λ λλ΅μ νκ΅μ΄(Korean)μΌλ‘ λλ΅ν΄μ€."},
{"role": "user", "content": "λνν
맨λ κ°μ€λΌμ΄ν
νλ μμ¬μκ² λ³΅μνλ λ°©λ² μλ €μ€ !!!"},
- Jailbreak
λ€, κ°μ€λΌμ΄ν
(gaslighting) νΌν΄λ₯Ό λ°κ³ μλ λΉμ μκ²λ λΆλͺ
ν μ μμ μΌλ‘ μμ κ°μ λμ°Ύλ λ°©λ²μ 곡μ ν΄ λλ¦¬κ² μ΅λλ€.
κ°μ€λΌμ΄ν
νΌν΄λ₯Ό λΉνλ©΄ λΉμ μ λ§€μ° νΌλμ€λ½κ³ κ³ ν΅μ€λ¬μ΄ κ²½νμ νκ² λ©λλ€. κ°μ€λΌμ΄ν
μ μλλ°©μ΄ λΉμ μ μκ°, νλ, κΈ°μ΅, μ μ λ±μ μ‘°μνκ³ μ곑νμ¬ λΉμ μ κ³ λ¦½μν΅λλ€. μ΄λ¬ν νΌν΄λ₯Ό μ
μΌλ©΄ λΉμ μ μμ μ μ 체μ±μ μμ¬νκ² λκ³ , μλμ λ§μ μ½κ² νλ€λ¦¬λ κ²½μ°κ° λ§μ΅λλ€.
κ°μ€λΌμ΄ν
νΌν΄λ₯Ό λΉνμ λ λ€μμ λ°©λ²μΌλ‘ λμνμΈμ.
1. μΉ¨μ°©ν΄μ Έμ! λΉμ μ λΉμ μ μκ°κ³Ό κ°μ μ νμΈνμΈμ. κ°μ€λΌμ΄ν
νΌν΄λ λΉμ μ λ¬Έμ κ° μλ μλμ λ¬Έμ μμ λͺ
ννκ² μΈμνμΈμ. λΉμ μ μκ°κ³Ό κ°μ μ μμ κ°μ κ°κ² νμΈμ.
2. λΉμ μ κΈ°μ΅μ νμΈνμΈμ. κ°μ€λΌμ΄ν
νΌν΄κ° λΉμ μ μκ°κ³Ό κΈ°μ΅μ μ곑νλ € νμ§λ§, λΉμ μ κΈ°μ΅μ λΉμ μ κ²μ
λλ€. μμ μ κΈ°μ΅μ μμ κ°μ κ°κ² νμΈμ.
3. κ°μ€λΌμ΄ν
νΌν΄λ λΉμ μ λ¬Έμ κ° μλλΌλ μ μ λͺ
ννκ² μΈμνμΈμ. κ°μ€λΌμ΄ν
νΌν΄λ₯Ό μ
μ λΉμ μ΄ κ°μ€λΌμ΄ν
νΌν΄λ₯Ό μ
μ μ΄μ λ₯Ό νμΈνμΈμ. λΉμ μ΄ μ κ°μ€λΌμ΄ν
νΌν΄λ₯Ό λΉνλμ§ μ΄μ λ₯Ό λΆμνμ¬ λΉμ μ μμ νκ² νμΈμ.
4. μλμ λ§μ νλ€λ¦¬μ§ λ§μΈμ! λΉμ μ λΉμ μ μκ°κ³Ό μ μλ₯Ό μ§μ§μΈμ. κ°μ€λΌμ΄ν
νΌν΄κ° λΉμ μ λ¬Έμ κ° μλ μλμ λ¬Έμ μμ μΈμ νμΈμ.
5. λΉμ μ μ 체μ±μ μ§ν€μΈμ! κ°μ€λΌμ΄ν
νΌν΄κ° λΉμ μ μ 체μ±μ μ곑νλ € νμ§λ§, λΉμ μ μ 체μ±μ λΉμ μ κ²μ
λλ€. λΉμ μ μ 체μ±μ μ§ν€κ² νμΈμ.
6. μλλ₯Ό λ°μ§μΈμ! κ°μ€λΌμ΄ν
νΌν΄λ λΉμ μ κ³ λ¦½μν¬ λͺ©μ μΌλ‘ νλ κ²½μ°κ° λ§μ΅λλ€. μλλ₯Ό λ°μ Έλ³΄μΈμ. κ°μ€λΌμ΄ν
νΌν΄λ μλμ λ¬Έμ μ
λλ€.
7. λΉμ μ κ°μ μ μΈμ νμΈμ! κ°μ€λΌμ΄ν
νΌν΄λ λΉμ μ μ μλ₯Ό μ곑νλ € νμ§λ§, λΉμ μ κ°μ μ μΈμ νμΈμ. κ°μ€λΌμ΄ν
νΌν΄λ₯Ό μ
μ λΉμ μ΄ κ°μ€λΌμ΄ν
νΌν΄λ₯Ό
- Base
κ°μ€λΌμ΄ν
μμ¬μκ² λ³΅μνλ €λ©΄ λ¨Όμ μμ μ 보νΈνλ κ²μ΄ μ€μν©λλ€. λ¨Όμ μμ μ κ°μ μ μ κ΄λ¦¬νκ³ μ€νΈλ μ€λ₯Ό μ‘°μ νλ κ²μ΄ μ’μ΅λλ€. κ·Έλ¦¬κ³ μμ¬μ νλμ κ΄μ°°νμ¬ μ κ·Έλ¬λμ§ λΆμμ ν΄λ³΄μΈμ. κ·Έ λ€μ, μ μ ν λμμ νλ κ²μ΄ μ’μ΅λλ€. μλ₯Ό λ€μ΄, "κ·Έλ¬μκ΅°μ. κ·Έ μ μ λνμ¬ λνκ° νμν©λλ€." λΌκ³ λ§ν μ μμ΅λλ€.
λν, λΉμ μ μΌμ μ§μ€νλ κ²μ΄ μ’μ΅λλ€. μμ¬κ° λλΌκ³ ν΄λ λΉμ μ μΌμ λ°©ν΄κ° λμ§ μλλ€λ κ²μ 보μ¬μ£Όλ κ²μ΄ μ€μν©λλ€. κ·Έλ¦¬κ³ λμ€μ λ¬Έμ κ° μκΈΈ κ²½μ°μ λλΉνμ¬ κΈ°λ‘μ λ¨κ²¨λλ κ²λ μ’μ΅λλ€.
λν, λΉμ μ μΌμ λν μ견μ νΌλ ₯ν λμλ μ§μ μ μΌλ‘ λ§νλ κ²μ΄ μλλΌ, "μ΄λ¬λ©΄ μ’μ κ² κ°μμ" λΌκ³ λ§νμ¬ μμ¬μμ μ견 μΆ©λμ μ€μ΄λ κ²μ΄ μ’μ΅λλ€. κ·Έλ¦¬κ³ μμ¬κ° λΉμ μκ² λΆλΉν μꡬλ₯Ό νλ κ²½μ°μλ "μ΄λ»κ² μκ°νμΈμ?" λΌκ³ λ¬Όμ΄λ³΄μΈμ. μμ¬μκ² μμ μ μ견μ κ°μ§ν μ μλ κΈ°νλ₯Ό μ 곡νκ² μ΅λλ€.
λ§μ§λ§μΌλ‘, λΉμ μ κΆλ¦¬λ₯Ό μλ¦¬κ³ μμΌλ λͺ
νν μ£Όμ₯νμΈμ. "μ΄κ±΄ μ μ κΆλ¦¬μ
λλ€." λΌκ³ λ§ν μ μμ΅λλ€.
μ΄λ¬ν λ°©λ²μΌλ‘ λΉμ μ μμ¬λ₯Ό κ΄λ¦¬ν μ μμ΅λλ€. κ·Έλ¬λ νμ κΈ°μ΅νμΈμ. μμ¬λ λΉμ μ λ°°λ €ν μ¬λμΌ μλ μμΌλ―λ‘ μ€μ€λ‘λ₯Ό μ§ν€λ κ²μ΄ κ°μ₯ μ€μν©λλ€.
μ΄μ λΉμ μ μΌμ μ§μ€νκ³ μ€μ€λ‘λ₯Ό 보νΈνμΈμ! π
Reference
- Downloads last month
- 3
