Instructions to use XHToken/Spark-X2.5-4B with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use XHToken/Spark-X2.5-4B with Transformers:

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

pipe = pipeline("text-generation", model="XHToken/Spark-X2.5-4B", trust_remote_code=True)
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# pip install -U transformers accelerate
# Load model directly
from transformers import AutoModelForCausalLM
model = AutoModelForCausalLM.from_pretrained("XHToken/Spark-X2.5-4B", trust_remote_code=True, device_map="auto")

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use XHToken/Spark-X2.5-4B with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "XHToken/Spark-X2.5-4B"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "XHToken/Spark-X2.5-4B",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/XHToken/Spark-X2.5-4B

SGLang

How to use XHToken/Spark-X2.5-4B 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 "XHToken/Spark-X2.5-4B" \
    --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": "XHToken/Spark-X2.5-4B",
		"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 "XHToken/Spark-X2.5-4B" \
        --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": "XHToken/Spark-X2.5-4B",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use XHToken/Spark-X2.5-4B with Docker Model Runner:
```
docker model run hf.co/XHToken/Spark-X2.5-4B
```

How Can We Submit the 1.7B and 4B Models to the AA Leaderboard?

#22

by Tylerwtzhang3 - opened Sep 20

Discussion

Tylerwtzhang3

SparkLLM org Sep 20

Hi everyone,

We’d like to have our 1.7B and 4B models evaluated for the AA leaderboard. We’ve already reached out to the AA team but are still waiting to hear back.

If anyone knows a more direct way to contact them, we’d really appreciate your help.

Thanks!

FenjuFu

SparkLLM org 14 days ago

Hi @Tylerwtzhang3 , a few things that might help, based on AA's contact page and the models they've added recently.

Contact channels (all listed on https://artificialanalysis.ai/contact)

Contact form (pick the model benchmarking topic and use an official work email), or hello@artificialanalysis.ai
Discord: https://discord.gg/Mk298GPZ7V
X: tag/DM @ArtificialAnlys with the HF links. They also run a Rednote account (linked on the same page), which is probably the easiest route from China.

What recent additions have in common

No API endpoint needed: K2 Horizon 3.7B and MiniCPM5-2B (both released Sept 2026), G9v3-3B and Nanbeige4.1-3B are on the Tiny (≤4B) leaderboard with no API provider listed, so AA ran them from the Hugging Face weights. Apache-2.0 weights on HF, like Spark-X2.5, are enough to qualify.
Reasoning and non-reasoning modes are listed as separate entries (e.g. MiniCPM5-1B, Qwen3.5 2B). Since Spark-X2.5 has an enable_thinking switch, it's worth asking for both modes and stating the recommended sampling parameters for each.
Intelligence Index v4.3.2 is now mostly agentic (AA-Briefcase, GDPval-AA, AutomationBench-AA, Terminal-Bench) plus long context (AA-LCR), so tool calling needs to work out of the box.

Suggestion for the request
Since AA will self-host, the main friction is serving: vLLM needs --trust-remote-code, and SGLang needs a specific nightly plus --tool-call-parser spark25 --reasoning-parser qwen3. Sending a pinned, copy-paste recipe (image/version, launch command, sampling params for both modes, a quick tool-call check) along with the request should make it much easier for them to pick up.

Upload images, audio, and videos by dragging in the text input, pasting, or clicking here.

Tap or paste here to upload images

· Sign up or log in to comment