Instructions to use officialmobai/Rostam-S1-4B with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use officialmobai/Rostam-S1-4B with Transformers:

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

pipe = pipeline("text-generation", model="officialmobai/Rostam-S1-4B", 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("officialmobai/Rostam-S1-4B", trust_remote_code=True, device_map="auto")

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use officialmobai/Rostam-S1-4B with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "officialmobai/Rostam-S1-4B"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "officialmobai/Rostam-S1-4B",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/officialmobai/Rostam-S1-4B

SGLang

How to use officialmobai/Rostam-S1-4B 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 "officialmobai/Rostam-S1-4B" \
    --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": "officialmobai/Rostam-S1-4B",
		"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 "officialmobai/Rostam-S1-4B" \
        --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": "officialmobai/Rostam-S1-4B",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use officialmobai/Rostam-S1-4B with Docker Model Runner:
```
docker model run hf.co/officialmobai/Rostam-S1-4B
```
Browse Quantizations to use this model in llama.cpp, Ollama, LM Studio, or any compatible app.

⚔️ Rostam-S1 (رستم)

Frontier Reasoning & Agentic Model by Mob AI (in collaboration with Spark)

Rostam-S1 is an elite intelligent reasoning and agentic model developed by Mob AI in strategic technological collaboration with the Spark project. Named after the legendary champion of Persian epic literature whose name personifies valor, strategic intellect, and unwavering perseverance, Rostam is engineered to deliver frontier capabilities across Mathematics, Theoretical Physics, Scientific Coding, Persian (Farsi) Cognition, and Agentic Systems.

📌 Development Philosophy: Architectural Excellence

Rostam is built on the conviction that architectural precision, high-order reasoning traces, and disciplined alignment triumph over brute-force scaling. Rather than defining capability by parameter volume, Rostam focuses on:

Verifiable Symbolic Reasoning: Step-by-step mathematical proofs and physical derivations.
Deep Cognitive Duality: Native bilingual mastery across English and Persian technical thought.
Zero Catastrophic Forgetting: Subspace-isolated parameter adaptation protected by persistent replay regularization.

🎯 Target Benchmarks & Alignment Focus

Note on Evaluation Transparency: The numbers in the Target column represent the mathematical optimization goals designed for the full post-training curriculum. Official audited benchmark scores will be released upon completion of the full RLVR evaluation pipeline.

Benchmark	Domain	Spark-X2.6 Baseline*	Competitor Average*	Rostam Target
GPQA Diamond	Physics & Hard Science	77.1%	80.2%	Target: >82%
AIME 2026	Olympiad Mathematics	91.2%	85.1%	Target: >92%
Gaokao 2026	Advanced Math	134.9 / 150	133.0 / 150	Target: >138
SciCode	Scientific Simulation	36.8%	36.2%	Target: >42%
SWE-bench Verified	Agentic Software Engineering	62.8%	48.6%	Target: >64%
Persian STEM & QA	Bilingual Technical Reasoning	Baseline	Untested	Target: Frontier

*Baseline and competitor numbers are from officially released model cards and papers.

🚀 Key Features

Dual-Channel Deep Reasoning (<think>): Native controllable reasoning traces providing step-by-step analytical derivations before delivering definitive conclusions.
Deep Persian (فارسی) Cognitive Integration: Full bilingual fluency across technical terminology, complex physics derivations, software engineering, and formal Persian prose.
Agentic Tool Integration: Native function calling and verifiable sandbox execution.

💻 Quickstart Inference

import torch
from transformers import AutoModelForCausalLM, AutoTokenizer

model_id = "officialmobai/Rostam-S1-4B"

tokenizer = AutoTokenizer.from_pretrained(model_id, trust_remote_code=True)
model = AutoModelForCausalLM.from_pretrained(
    model_id,
    trust_remote_code=True,
    dtype=torch.bfloat16,
    device_map="auto"
)

messages = [
    {"role": "user", "content": "نام شما چیست و سازنده شما کیست؟"}
]

inputs = tokenizer.apply_chat_template(
    messages,
    tokenize=True,
    add_generation_prompt=True,
    enable_thinking=True,
    return_dict=True,
    return_tensors="pt"
).to(model.device)

outputs = model.generate(
    **inputs,
    max_new_tokens=512,
    do_sample=True,
    temperature=0.7,
    top_p=0.9
)

print(tokenizer.decode(outputs[0][inputs["input_ids"].shape[-1]:], skip_special_tokens=True))