Instructions to use roskosmos19/Orca-4B-flash-max with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use roskosmos19/Orca-4B-flash-max with Transformers:

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

pipe = pipeline("text-generation", model="roskosmos19/Orca-4B-flash-max")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("roskosmos19/Orca-4B-flash-max")
model = AutoModelForCausalLM.from_pretrained("roskosmos19/Orca-4B-flash-max", 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 roskosmos19/Orca-4B-flash-max with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "roskosmos19/Orca-4B-flash-max"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "roskosmos19/Orca-4B-flash-max",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/roskosmos19/Orca-4B-flash-max

SGLang

How to use roskosmos19/Orca-4B-flash-max 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 "roskosmos19/Orca-4B-flash-max" \
    --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": "roskosmos19/Orca-4B-flash-max",
		"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 "roskosmos19/Orca-4B-flash-max" \
        --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": "roskosmos19/Orca-4B-flash-max",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use roskosmos19/Orca-4B-flash-max with Docker Model Runner:
```
docker model run hf.co/roskosmos19/Orca-4B-flash-max
```

A newer version of this model is available: roskosmos19/Orca-4B-Instruct

SystemOne-4B-Agentic-AGI

A 4-billion-parameter System One style decision & agentic model
Focused on fast, calibrated, typed decisions + stronger AGI-oriented agentic capabilities.

This repository contains the non-weight files (configs, tokenizer, templates, license, documentation) for a conceptual / research System One model.
No model.safetensors files are included (as requested).

What is a System One Model?

Inspired by TypeSafe AI’s Jev and Kahneman’s System 1 thinking:

Takes unstructured state + typed questions
Returns typed answers with calibrated probabilities (Choice / Score / Noul)
Parallel evaluation of multiple questions
No free-form text generation for decisions → zero schema / type errors by construction
Extremely low latency and cost for decision workloads
Designed to be composed inside software (smart if-statements, routing, scoring, guardrails)

This 4B variant adds stronger agentic / AGI-oriented capabilities:

Tool-use and multi-step planning
Self-reflection and confidence-aware routing
Better long-horizon agent behavior
Improved calibration under uncertainty

Model Details

Property	Value
Architecture	Qwen3.5-based (dense)
Parameters	~4B
Context length	32k – 64k (depending on config)
Primary focus	System One decisions + Agentic
Training objective	RLCD-style + agentic trajectories
License	Apache 2.0
Base	Qwen/Qwen3.5-4B

Files in this repository

config.json – model configuration
tokenizer.json / tokenizer_config.json / vocab.json / merges.txt – tokenizer
chat_template.jinja – chat / system-one template
preprocessor_config.json / video_preprocessor_config.json – preprocessor configs
LICENSE – Apache 2.0
.gitattributes – Git LFS attributes
README.md – this file

Intended Use

High-throughput decision making inside applications
Agentic workflows that need reliable typed outputs + confidence
Research into System One / Machine-Native Intelligence
Prototyping AGI-style agents that combine fast decisions with deeper reasoning

Limitations

This package does not include the actual model weights.
Calibration quality and agentic performance depend on the final trained checkpoint.
English is the strongest language; other languages may require additional fine-tuning.

Citation / Inspiration

TypeSafe AI – System One Models & Jev (2026)
Qwen team – Qwen3.5 series
Kahneman – Thinking, Fast and Slow

How to use (once weights are available)

from transformers import AutoModelForCausalLM, AutoTokenizer

model = AutoModelForCausalLM.from_pretrained("your-org/SystemOne-4B-Agentic-AGI", trust_remote_code=True)
tokenizer = AutoTokenizer.from_pretrained("your-org/SystemOne-4B-Agentic-AGI")

# Example System One style call would go through a custom head or constrained decoding

Created for research and experimentation with System One + Agentic AGI paradigms.
No model weights included.

Downloads last month: 517

Safetensors

Model size

5B params

Tensor type

BF16

F32