Instructions to use game-plus/qwen3.5-4b-gameworld-sft-final with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use game-plus/qwen3.5-4b-gameworld-sft-final with Transformers:

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

pipe = pipeline("image-text-to-text", model="game-plus/qwen3.5-4b-gameworld-sft-final")
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)

# pip install -U transformers accelerate
# Load model directly
from transformers import AutoProcessor, AutoModelForMultimodalLM

processor = AutoProcessor.from_pretrained("game-plus/qwen3.5-4b-gameworld-sft-final")
model = AutoModelForMultimodalLM.from_pretrained("game-plus/qwen3.5-4b-gameworld-sft-final", device_map="auto")
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=256)
print(processor.decode(outputs[0][inputs["input_ids"].shape[-1]:]))

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use game-plus/qwen3.5-4b-gameworld-sft-final with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "game-plus/qwen3.5-4b-gameworld-sft-final"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "game-plus/qwen3.5-4b-gameworld-sft-final",
		"messages": [
			{
				"role": "user",
				"content": [
					{
						"type": "text",
						"text": "Describe this image in one sentence."
					},
					{
						"type": "image_url",
						"image_url": {
							"url": "https://cdn.britannica.com/61/93061-050-99147DCE/Statue-of-Liberty-Island-New-York-Bay.jpg"
						}
					}
				]
			}
		]
	}'

Use Docker

docker model run hf.co/game-plus/qwen3.5-4b-gameworld-sft-final

SGLang

How to use game-plus/qwen3.5-4b-gameworld-sft-final 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 "game-plus/qwen3.5-4b-gameworld-sft-final" \
    --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": "game-plus/qwen3.5-4b-gameworld-sft-final",
		"messages": [
			{
				"role": "user",
				"content": [
					{
						"type": "text",
						"text": "Describe this image in one sentence."
					},
					{
						"type": "image_url",
						"image_url": {
							"url": "https://cdn.britannica.com/61/93061-050-99147DCE/Statue-of-Liberty-Island-New-York-Bay.jpg"
						}
					}
				]
			}
		]
	}'

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 "game-plus/qwen3.5-4b-gameworld-sft-final" \
        --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": "game-plus/qwen3.5-4b-gameworld-sft-final",
		"messages": [
			{
				"role": "user",
				"content": [
					{
						"type": "text",
						"text": "Describe this image in one sentence."
					},
					{
						"type": "image_url",
						"image_url": {
							"url": "https://cdn.britannica.com/61/93061-050-99147DCE/Statue-of-Liberty-Island-New-York-Bay.jpg"
						}
					}
				]
			}
		]
	}'

Docker Model Runner
How to use game-plus/qwen3.5-4b-gameworld-sft-final with Docker Model Runner:
```
docker model run hf.co/game-plus/qwen3.5-4b-gameworld-sft-final
```

You need to agree to share your Hugging Face information to access this model

This repository is publicly accessible, but you have to accept the conditions to access its files and content.

Qwen3.5-4B GameWorld SFT — final (1 epoch, step 9237)

Full-parameter SFT of Qwen/Qwen3.5-4B (language model trained; vision encoder and aligner frozen) on the GameWorld game-agent data. Final checkpoint after one full epoch (9237 steps). Final train loss 0.283 (running average over the epoch). (running average over the epoch; last logged step 0.257).

Training setup

Data: 5 sources mixed (4 teacher-model sources + policy rollouts), 1,564,236 rows, ~10.9B tokens, max 25,000 tokens/sample. Each sample: system prompt, user turn with up to 7 game screenshots (≤1024 image tokens each), assistant <think>\n\n</think> + <tool_call> action. Policy samples mask the empty-think turn from the loss; teacher samples train on the whole assistant message.
Optimizer: AdamW, lr 1e-5 cosine, warmup 3%, per-device batch 2 × accumulation 3 = 48 packed sequences × 25K tokens ≈ 1.18M tokens/step (~165 samples).
BF16, DeepSpeed ZeRO-2, sequence packing, Liger kernel, flash-attention 2, gradient checkpointing.
1 node × 8 H100 80GB, ~37 h wall clock, 14.3 s/step. ms-swift 4.5.3, transformers 5.16.1, DeepSpeed 0.19.8.
Weights: bf16 safetensors, 3 shards (~9.3 GB).

Inference

Set IMAGE_MAX_TOKEN_NUM=1024 (or the equivalent max_pixels) so screenshots are tokenized as in training.

from transformers import AutoProcessor, AutoModelForImageTextToText
import torch
repo = "game-plus/qwen3.5-4b-gameworld-sft-final"
processor = AutoProcessor.from_pretrained(repo)
model = AutoModelForImageTextToText.from_pretrained(repo, dtype=torch.bfloat16, device_map="cuda")
messages = [{"role": "system", "content": SYSTEM_PROMPT},
            {"role": "user", "content": [{"type": "image", "image": "screen.png"}, {"type": "text", "text": USER_TEXT}]}]
inputs = processor.apply_chat_template(messages, add_generation_prompt=True, tokenize=True, return_dict=True, return_tensors="pt").to(model.device)
out = model.generate(**inputs, max_new_tokens=256)
print(processor.decode(out[0][inputs["input_ids"].shape[1]:], skip_special_tokens=True))

With ms-swift: swift infer --model game-plus/qwen3.5-4b-gameworld-sft-final --template qwen3_5 --use_hf true. The model emits an empty think block followed by a <tool_call> JSON, matching the training format.

Downloads last month: 2

Safetensors

Model size

5B params

Tensor type

BF16

Model tree for game-plus/qwen3.5-4b-gameworld-sft-final

Base model

Qwen/Qwen3.5-4B-Base

Finetuned

Qwen/Qwen3.5-4B

Finetuned

(1108)

this model