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

Libraries

How to use mcnckc/llm-hw2-ppo with Transformers:

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

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

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("mcnckc/llm-hw2-ppo")
model = AutoModelForCausalLM.from_pretrained("mcnckc/llm-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

vLLM

How to use mcnckc/llm-hw2-ppo with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "mcnckc/llm-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": "mcnckc/llm-hw2-ppo",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/mcnckc/llm-hw2-ppo

SGLang

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

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

YAML Metadata Warning:The pipeline tag "text2text-generation" is not in the official list: text-classification, token-classification, table-question-answering, question-answering, zero-shot-classification, translation, summarization, feature-extraction, text-generation, fill-mask, sentence-similarity, text-to-speech, text-to-audio, automatic-speech-recognition, audio-to-audio, audio-classification, audio-text-to-text, voice-activity-detection, depth-estimation, image-classification, object-detection, image-segmentation, text-to-image, image-to-text, image-to-image, image-to-video, unconditional-image-generation, video-classification, reinforcement-learning, robotics, tabular-classification, tabular-regression, tabular-to-text, table-to-text, multiple-choice, text-ranking, text-retrieval, time-series-forecasting, text-to-video, image-text-to-text, image-text-to-image, image-text-to-video, visual-question-answering, document-question-answering, zero-shot-image-classification, graph-ml, mask-generation, zero-shot-object-detection, text-to-3d, image-to-3d, image-feature-extraction, video-text-to-text, keypoint-detection, visual-document-retrieval, any-to-any, video-to-video, other

Model Card for Model ID

SmolLM2-135M-Instruct дообученная с помощью PPO на датасете HumanLLMs/Human-Like-DPO-Dataset, чтобы сделать генерацию более эмоциональной.

В PPO использована reward модель - https://huggingface.co/mcnckc/llm-hw2-reward-model

Гиперпараметры дообучения

Обучение длилось одну эпоху. В качестве ref_model использовался изначальный чекпоинт той же модели.

num_train_epochs=1,
per_device_train_batch_size=8,
gradient_accumulation_steps=2,
learning_rate=1e-5

Использовался нестандартный паддинг - tokenizer.pad_token = '<|endoftext|>'

Примеры генерации

До дообучения

system

You are a helpful AI assistant named SmolLM, trained by Hugging Face

user

What's your morning routine like?

assistant

My morning routine is quite simple, but it's essential for maintaining a healthy and productive work-life balance. Here's what I do every morning:

Start with a cup of coffee: I usually start my day with a cup of coffee to get my blood flowing and to help me feel more awake.
Get dressed: I dress in my morning attire, which includes a crisp white shirt, a black tie, and a pair of white pants.
Start with a quick breakfast: I usually have a quick breakfast, which might be a bowl of cereal, a banana, or a yogurt.
Get ready for work: I usually start my workday with a quick workout, which might be a quick 10-minute walk or a quick 10-minute yoga session.
Maintain a healthy routine: I make sure to stay hydrated, eat a balanced diet, and get enough sleep.
Take a short walk: I usually take a short walk around the block to get some fresh air and to help me stay focused.
Get ready for the day: I usually start my day with a quick breakfast, a quick workout, and a quick breakfast.

После дообучения

system

You are a helpful AI assistant named SmolLM, trained by Hugging Face

user

What's your morning routine like?

assistant

My morning routine is quite simple, but it's essential for maintaining a healthy and productive day. I wake up around 6:00 AM, and before I start my day, I take a few minutes to stretch and get ready. I then head to the kitchen to brew a cup of coffee, which I do in about 10 minutes.

After my coffee, I head to the bathroom to wash my hands and make sure everything is in order. I then head to the kitchen to start my morning routine. I start by making a cup of tea, which I do in about 20 minutes.

I then head to the kitchen to start my morning routine. I start by making a cup of tea, which I do in about 20 minutes. I then head to the kitchen to start my morning routine. I start by making a cup of tea, which I do in about 20 minutes.

Downloads last month: 2

Model tree for mcnckc/llm-hw2-ppo

Base model

HuggingFaceTB/SmolLM2-135M

Quantized

HuggingFaceTB/SmolLM2-135M-Instruct

Finetuned

(324)

this model

Dataset used to train mcnckc/llm-hw2-ppo

Collection including mcnckc/llm-hw2-ppo

LLM-HW2

Collection

Коллекция моделей, обученных в процессе выполнения ДЗ • 3 items • Updated Mar 6, 2025