Instructions to use Losa10/Qwen3-4B-Thinking-Slerp-F with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use Losa10/Qwen3-4B-Thinking-Slerp-F with Transformers:

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

pipe = pipeline("text-generation", model="Losa10/Qwen3-4B-Thinking-Slerp-F")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("Losa10/Qwen3-4B-Thinking-Slerp-F")
model = AutoModelForCausalLM.from_pretrained("Losa10/Qwen3-4B-Thinking-Slerp-F", 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 Losa10/Qwen3-4B-Thinking-Slerp-F with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "Losa10/Qwen3-4B-Thinking-Slerp-F"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "Losa10/Qwen3-4B-Thinking-Slerp-F",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/Losa10/Qwen3-4B-Thinking-Slerp-F

SGLang

How to use Losa10/Qwen3-4B-Thinking-Slerp-F 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 "Losa10/Qwen3-4B-Thinking-Slerp-F" \
    --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": "Losa10/Qwen3-4B-Thinking-Slerp-F",
		"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 "Losa10/Qwen3-4B-Thinking-Slerp-F" \
        --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": "Losa10/Qwen3-4B-Thinking-Slerp-F",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use Losa10/Qwen3-4B-Thinking-Slerp-F with Docker Model Runner:
```
docker model run hf.co/Losa10/Qwen3-4B-Thinking-Slerp-F
```
Browse Quantizations to use this model in llama.cpp, Ollama, LM Studio, or any compatible app.

Qwen3-4B-Thinking-Slerp-FH

This model is a Spherical Linear Interpolation (SLERP) merge of two specialized Qwen3-4B variants. It aims to combine high-level reasoning capabilities from distilled thinking models with refined instruction-following from SFT (Supervised Fine-Tuning).

Merge Details

Merge Method

This model was merged using the SLERP (Spherical Linear Interpolation) method. SLERP is ideal for merging models of the same architecture as it preserves the geometric characteristics of the weight space better than traditional linear averaging.

Models Merged

The following models were included in the merge:

TeichAI/Qwen3-4B-Thinking...Distill: High-reasoning capabilities via distillation.
Losa10/Qwen3-4B-Thinking...SFT: Enhanced thinking and chat instruction following.

Configuration

The merge used a dynamic interpolation factor ($t$) to balance attention and MLP weights across the model layers:

slices:
  - sources:
      - model: TeichAI/Qwen3-4B-Thinking-2507-Gemini-3-Pro-Preview-High-Reasoning-Distill
        layer_range: [0, 36]
      - model: Losa10/Qwen3-4B-Thinking-2507-G3P-F-SFT
        layer_range: [0, 36]
merge_method: slerp
base_model: TeichAI/Qwen3-4B-Thinking-2507-Gemini-3-Pro-Preview-High-Reasoning-Distill
parameters:
  t:
    - filter: self_attn
      value: [0.4, 0.6] # Favors SFT in deeper attention layers
    - filter: mlp
      value: [0.6, 0.4] # Favors Reasoning in deeper MLP layers
    - value: 0.5
dtype: bfloat16

Downloads last month: 7

Safetensors

Model size

4B params

Tensor type

BF16

Model tree for Losa10/Qwen3-4B-Thinking-Slerp-F

Losa10/Qwen3-4B-Thinking-2507-G3P-F-SFT

TeichAI/Qwen3-4B-Thinking-2507-Gemini-3-Pro-Preview-High-Reasoning-Distill

Merge model

this model

Quantizations

1 model