Instructions to use SamsungSAILMontreal/Qwen3-4B-Instruct-2507-Code with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use SamsungSAILMontreal/Qwen3-4B-Instruct-2507-Code with Transformers:

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

pipe = pipeline("text-generation", model="SamsungSAILMontreal/Qwen3-4B-Instruct-2507-Code")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("SamsungSAILMontreal/Qwen3-4B-Instruct-2507-Code")
model = AutoModelForCausalLM.from_pretrained("SamsungSAILMontreal/Qwen3-4B-Instruct-2507-Code", device_map="auto")
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 SamsungSAILMontreal/Qwen3-4B-Instruct-2507-Code with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "SamsungSAILMontreal/Qwen3-4B-Instruct-2507-Code"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "SamsungSAILMontreal/Qwen3-4B-Instruct-2507-Code",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/SamsungSAILMontreal/Qwen3-4B-Instruct-2507-Code

SGLang

How to use SamsungSAILMontreal/Qwen3-4B-Instruct-2507-Code 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 "SamsungSAILMontreal/Qwen3-4B-Instruct-2507-Code" \
    --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": "SamsungSAILMontreal/Qwen3-4B-Instruct-2507-Code",
		"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 "SamsungSAILMontreal/Qwen3-4B-Instruct-2507-Code" \
        --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": "SamsungSAILMontreal/Qwen3-4B-Instruct-2507-Code",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use SamsungSAILMontreal/Qwen3-4B-Instruct-2507-Code with Docker Model Runner:
```
docker model run hf.co/SamsungSAILMontreal/Qwen3-4B-Instruct-2507-Code
```

Qwen3-4B-Instruct-2507-Code

This model is obtained by fine-tuning Qwen/Qwen3-4B-Instruct-2507 on the evol-codealpaca-v1 train split. The model is used in the experiments described in https://bknyaz.github.io/blog/2026/meta-merge/. Single A100 was used for fine-tuning and evaluation.

The following versions were used for train/eval:

python >= 3.10
torch : 2.9.0+cu128
lm_eval : 0.4.9.1
vllm : 0.11.1
transformers : 4.57.6
datasets : 3.2.0
numpy : 2.2.6

Training

The TRL library was used with SFT/full-rank options:

python trl/scripts/sft.py --model_name_or_path Qwen/Qwen3-4B-Instruct-2507 --dataset_name theblackcat102/evol-codealpaca-v1 --learning_rate 2e-5 \
--num_train_epochs 1 --per_device_train_batch_size 2 --gradient_accumulation_steps 8 --gradient_checkpointing --eos_token '<|im_end|>' --eval_strategy no \
--completion_only_loss True --report_to wandb --output_dir /path/to/the/finetuned/model

This is by far not the most compute and performance efficient fine-tuning, but it could be a good baseline.

The dataset was preprocessed to the conversational format:

# trl/scripts/sft.py

dataset = load_dataset(...)

def preprocess_function(example):
  return {
  "prompt": [{"role": "user", "content": example["instruction"]}],
  "completion": [
      {"role": "assistant", "content": example['output']}
  ],
  }

dataset = dataset.map(preprocess_function)

Evaluation

Evaluation was done with lm_eval on the humaneval (instruct) benchmark:

python -m lm_eval --model vllm --model_args pretrained=${model},tensor_parallel_size=1,dtype=auto,gpu_memory_utilization=0.9,data_parallel_size=1 \
 --tasks humaneval_instruct --batch_size 1 --apply_chat_template=True --confirm_run_unsafe_code --trust_remote_code