Instructions to use benjamin/Gemma2-2B-IT-with-Qwen2-Tokenizer with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use benjamin/Gemma2-2B-IT-with-Qwen2-Tokenizer with Transformers:

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

pipe = pipeline("text-generation", model="benjamin/Gemma2-2B-IT-with-Qwen2-Tokenizer")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("benjamin/Gemma2-2B-IT-with-Qwen2-Tokenizer")
model = AutoModelForCausalLM.from_pretrained("benjamin/Gemma2-2B-IT-with-Qwen2-Tokenizer")
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 benjamin/Gemma2-2B-IT-with-Qwen2-Tokenizer with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "benjamin/Gemma2-2B-IT-with-Qwen2-Tokenizer"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "benjamin/Gemma2-2B-IT-with-Qwen2-Tokenizer",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/benjamin/Gemma2-2B-IT-with-Qwen2-Tokenizer

SGLang

How to use benjamin/Gemma2-2B-IT-with-Qwen2-Tokenizer 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 "benjamin/Gemma2-2B-IT-with-Qwen2-Tokenizer" \
    --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": "benjamin/Gemma2-2B-IT-with-Qwen2-Tokenizer",
		"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 "benjamin/Gemma2-2B-IT-with-Qwen2-Tokenizer" \
        --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": "benjamin/Gemma2-2B-IT-with-Qwen2-Tokenizer",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use benjamin/Gemma2-2B-IT-with-Qwen2-Tokenizer with Docker Model Runner:
```
docker model run hf.co/benjamin/Gemma2-2B-IT-with-Qwen2-Tokenizer
```

Gemma2-2B Instruction Tuned Model (Transferred to Qwen Tokenizer) Model Card

Gemma2-2B-IT transferred to the Qwen2 Tokenizer. The model approximately preserves performance of the original on most benchmarks, except for some slight degradations.

Model Details

Base Model: Gemma2-2B
Tokenization: Transferred to the Qwen Tokenizer
Training Methodology: Instruction-tuned Gemma2-2B-IT transferred to the Qwen Tokenizer

Benchmark	Gemma2-2B w/ Qwen Tokenizer	Original Gemma2-2B-IT
PiQA	76.9	79.6
HS	70.7	72.5
ARC-C	46.8	50.4
BoolQ	82.8	83.8
MMLU	53.8	56.9
Arith.	83.9	84.8
IFEval	62.5	62.5

Model Details

Details on the training methodology are forthcoming.

Use

import torch
from transformers import pipeline

pipe = pipeline(
    "text-generation",
    model="benjamin/Gemma2-2B-IT-with-Qwen2-Tokenizer",
    model_kwargs={"torch_dtype": torch.bfloat16},
    device="cuda",  # replace with "mps" to run on a Mac device
)

messages = [
    {"role": "user", "content": "Who are you? Please, answer in pirate-speak."},
]

outputs = pipe(messages, max_new_tokens=256)
assistant_response = outputs[0]["generated_text"][-1]["content"].strip()
print(assistant_response)