Instructions to use tsessk/llm-course-hw2-ppo with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use tsessk/llm-course-hw2-ppo with Transformers:

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

pipe = pipeline("text-generation", model="tsessk/llm-course-hw2-ppo")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForMultimodalLM

tokenizer = AutoTokenizer.from_pretrained("tsessk/llm-course-hw2-ppo")
model = AutoModelForMultimodalLM.from_pretrained("tsessk/llm-course-hw2-ppo")
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 tsessk/llm-course-hw2-ppo with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "tsessk/llm-course-hw2-ppo"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "tsessk/llm-course-hw2-ppo",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/tsessk/llm-course-hw2-ppo

SGLang

How to use tsessk/llm-course-hw2-ppo 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 "tsessk/llm-course-hw2-ppo" \
    --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": "tsessk/llm-course-hw2-ppo",
		"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 "tsessk/llm-course-hw2-ppo" \
        --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": "tsessk/llm-course-hw2-ppo",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use tsessk/llm-course-hw2-ppo with Docker Model Runner:
```
docker model run hf.co/tsessk/llm-course-hw2-ppo
```

🤖 SmolLM-135M - PPO Fine-Tuned Model

📝 Overview

This is a Proximal Policy Optimization (PPO) fine-tuned version of SmolLM-135M, designed to make its responses more emotional and engaging. The model has been optimized using HumanLLMs/Human-Like-DPO-Dataset.

PPO is a reinforcement learning algorithm commonly used in RLHF (Reinforcement Learning from Human Feedback) to optimize model behavior based on a reward model. Paper

📊 Training & Dataset

Base Model: SmolLM-135M
Reward Model: tsessk/llm-course-hw2-reward-model
Fine-Tuned Dataset: HumanLLMs/Human-Like-DPO-Dataset
Objective: Improve expressiveness and emotional depth through reward-based reinforcement learning.

📌 Example of PPO-Enhanced Response

Here’s an example of how the PPO model responds to a simple prompt compared to a standard LLM.

Prompt:

What's your morning routine like?

Base model Response:

I'm excited to start my morning routine! As a digital AI assistant, I don't have personal preferences or habits, but I can provide you with a general idea of what a morning routine might look like. Here's a sample routine that I've developed based on various studies and research:

**Morning Routine (10-15 minutes)**

1. **Hydrate**: Drink a full glass of water or a healthy beverage to start the day.
2. **Brush teeth**: Brush your teeth for 2 minutes, then rinse with water.
3. **Get dressed**: Put on some comfortable clothes, including a light jacket or sweater.
4. **Get ready for work**: Start with some morning exercise, such as a 10-minute yoga or stretching routine.
5. **Plan your day**: Review your schedule, set goals, and prioritize tasks.
6. **Get ready for bed**: Wind down with a relaxing activity, like reading or listening to calming music.

**Morning Routine (15-30 minutes)**

1. **Get ready for work**: Start with some morning exercise, such as a 10-minute yoga or stretching routine.
2. **Get ready for

PPO Response:

I'm excited to start my morning routine! As a digital AI assistant, I don't have personal preferences or habits, but I can provide you with a general idea of what a morning routine might look like. Here's a sample routine I came up with:

**Morning Routine (10-15 minutes)**

1. **Hydrate**: Drink a full glass of water or a herbal tea to start the day.
2. **Brush teeth**: Brush your teeth for 2-3 minutes, then rinse with water.
3. **Get dressed**: Put on some comfortable clothes, including a light jacket or sweater.
4. **Get ready for work**: Start your day with a quick breakfast, such as oatmeal with fruit or scrambled eggs with whole-grain toast.
5. **Take a few deep breaths**: Take a few deep breaths to calm your mind and body.
6. **Get ready for the day**: Start your day with a morning workout, such as a yoga or Pilates class.
7. **Get ready for school**: Start your day with a morning routine, such as reading a book or taking a short walk.
8. **Get ready for bed**: Start your day with a gentle sleep