Instructions to use pavelfedortsov/gemma4-e2b-colloquial-ru-merged with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use pavelfedortsov/gemma4-e2b-colloquial-ru-merged with Transformers:

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

pipe = pipeline("text-generation", model="pavelfedortsov/gemma4-e2b-colloquial-ru-merged")
messages = [
    {
        "role": "user",
        "content": [
            {"type": "image", "url": "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/p-blog/candy.JPG"},
            {"type": "text", "text": "What animal is on the candy?"}
        ]
    },
]
pipe(text=messages)

# Load model directly
from transformers import AutoProcessor, AutoModelForMultimodalLM

processor = AutoProcessor.from_pretrained("pavelfedortsov/gemma4-e2b-colloquial-ru-merged")
model = AutoModelForMultimodalLM.from_pretrained("pavelfedortsov/gemma4-e2b-colloquial-ru-merged")
messages = [
    {
        "role": "user",
        "content": [
            {"type": "image", "url": "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/p-blog/candy.JPG"},
            {"type": "text", "text": "What animal is on the candy?"}
        ]
    },
]
inputs = processor.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(processor.decode(outputs[0][inputs["input_ids"].shape[-1]:]))

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use pavelfedortsov/gemma4-e2b-colloquial-ru-merged with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "pavelfedortsov/gemma4-e2b-colloquial-ru-merged"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "pavelfedortsov/gemma4-e2b-colloquial-ru-merged",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/pavelfedortsov/gemma4-e2b-colloquial-ru-merged

SGLang

How to use pavelfedortsov/gemma4-e2b-colloquial-ru-merged 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 "pavelfedortsov/gemma4-e2b-colloquial-ru-merged" \
    --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": "pavelfedortsov/gemma4-e2b-colloquial-ru-merged",
		"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 "pavelfedortsov/gemma4-e2b-colloquial-ru-merged" \
        --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": "pavelfedortsov/gemma4-e2b-colloquial-ru-merged",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use pavelfedortsov/gemma4-e2b-colloquial-ru-merged with Docker Model Runner:
```
docker model run hf.co/pavelfedortsov/gemma4-e2b-colloquial-ru-merged
```

gemma4-e2b-colloquial-ru-merged

English: Full-weight checkpoint: google/gemma-4-E2B-it merged with the colloquial Russian LoRA adapter for vLLM / RunPod deployment (no PEFT at inference time).

Что это

Полные веса = базовая модель google/gemma-4-E2B-it + LoRA gemma4-e2b-lora-colloquial-ru, объединённые для инференса на GPU (vLLM, RunPod Serverless).

Задача

Переписать формальный русский текст в разговорный стиль без мата, сохраняя факты, имена, цифры и структуру (абзацы, списки).

Обучение

~10k пар SFT (Telegram + social corpus), смешанный корпус
LoRA на language tower (r=16, alpha=16), затем merge в полные веса
Чекпоинт дополнен k_norm для слоёв 15–34 (совместимость с vLLM)

Использование

vLLM (RunPod Serverless)

MODEL_NAME=pavelfedortsov/gemma4-e2b-colloquial-ru-merged
LANGUAGE_MODEL_ONLY=true

См. docs/runpod_serverless_merged.md в исходном репозитории.

OpenAI-совместимый API (локальный proxy)

from openai import OpenAI

client = OpenAI(api_key="...", base_url="http://localhost:8080/v1")
r = client.chat.completions.create(
    model="colloquial-proxy",
    messages=[{"role": "user", "content": "Перепиши разговорным стилем:\n---\nТекст...\n---"}],
    max_tokens=512,
)
print(r.choices[0].message.content)

Streamlit UI

docker compose up → http://localhost:8501 (сервис Arteus Humanize).

Ограничения

Лицензия Gemma
Не предназначено для продакшена без собственной оценки качества и безопасности
Merged и LoRA-инференс могут слегка отличаться по стилю

Ссылки

Ресурс	URL
Base model	https://huggingface.co/google/gemma-4-E2B-it
LoRA adapter	https://huggingface.co/pavelfedortsov/gemma4-e2b-lora-colloquial-ru
RunPod deploy doc	см. репозиторий проекта `docs/runpod_serverless_merged.md`

Downloads last month: 305

Safetensors

Model size

5B params

Tensor type

F32

BF16

Model tree for pavelfedortsov/gemma4-e2b-colloquial-ru-merged

Base model

google/gemma-4-E2B

Finetuned

google/gemma-4-E2B-it

Finetuned

(232)

this model