teolm30

Ult2-Max

README

License: apache-2.0

Overview

Ult2 Max is designed with one core philosophy: maximum capability on minimum resources. While it leverages knowledge distilled from the most powerful AI models available, it is engineered to run efficiently on hardware that other models would struggle with.

Key Features

Expert-Level Intelligence: Trained on advanced reasoning, coding, math, science, and general knowledge
Ultra-Efficient: Only ~3B parameters - runs on CPUs, laptops, and basic hardware
Hardcoded System Prompt: Expert persona is built directly into the model configuration
Versatile: Handles programming, analysis, creative writing, math, and more
Safe & Responsible: Built with safety guidelines and responsible AI practices

Model Details

Table with columns: Property, Value
Property	Value
Base Model	Qwen/Qwen2.5-3B-Instruct
Parameters	~3.09B
Architecture	Qwen2ForCausalLM
Context Length	32,768 tokens
Language	English
License	Apache 2.0

GGUF Downloads (Recommended for Local Use)

GGUF files are optimized for local inference with llama.cpp, Ollama, and other compatible runtimes. This is the recommended format for running Ult2 Max on your own hardware.

Table with columns: Quantization, Size, Description
Quantization	Size	Description
Q4_K_M	~1.83 GB	Recommended - Best balance of quality and size. Fits in ~2GB RAM.
F16	~6.18 GB	Full precision - Maximum quality, requires ~8GB RAM.

Running with llama.cpp

bash
# Clone and build llama.cpp (or download pre-built binary)
git clone https://github.com/ggerganov/llama.cpp
cd llama.cpp && make

# Download the Q4_K_M GGUF
curl -LO https://huggingface.co/teolm30/Ult2-Max/resolve/main/Ult2-Max-Q4_K_M.gguf

# Run inference
./llama-cli -m Ult2-Max-Q4_K_M.gguf -p "What is quantum computing?" -n 256

Running with Ollama

Download the Q4_K_M GGUF:

bash
curl -LO https://huggingface.co/teolm30/Ult2-Max/resolve/main/Ult2-Max-Q4_K_M.gguf

Create a Modelfile:

markdown
FROM ./Ult2-Max-Q4_K_M.gguf
SYSTEM "You are Ult2 Max, an expert AI assistant."

Create and run:

bash
ollama create ult2max -f Modelfile
ollama run ult2max

Running with Python (Transformers)

python
from transformers import AutoModelForCausalLM, AutoTokenizer

model = AutoModelForCausalLM.from_pretrained(
    "teolm30/Ult2-Max",
    torch_dtype="auto",
    device_map="auto"
)
tokenizer = AutoTokenizer.from_pretrained("teolm30/Ult2-Max")

messages = [
    {"role": "user", "content": "Explain quantum computing in simple terms."}
]

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

outputs = model.generate(**inputs, max_new_tokens=512)
response = tokenizer.decode(outputs[0], skip_special_tokens=True)
print(response)

Hardware Requirements

Table with columns: Format, Minimum RAM, Recommended RAM
Format	Minimum RAM	Recommended RAM
Q4_K_M GGUF (CPU)	2 GB	4 GB
Q4_K_M GGUF (GPU)	2 GB VRAM	4 GB VRAM
F16 GGUF	8 GB	16 GB
SafeTensors (Transformers)	4 GB	8 GB GPU

Training & Development

Ult2 Max is built through a combination of techniques:

Base Model: Start with Qwen2.5-3B-Instruct, a highly capable small language model
System Prompt Integration: Expert-level system prompt hardcoded into the model configuration
Knowledge Distillation: Training data derived from analysis of advanced AI systems
Optimization: Model optimized for efficiency while maintaining quality

Capabilities

Programming: Write, debug, and explain code in multiple languages
Mathematics: Solve equations, explain concepts, step-by-step reasoning
Science: Explain complex scientific concepts clearly
Creative Writing: Stories, poetry, scripts, and more
Analysis: Data analysis, research, and problem-solving
General Knowledge: Factual Q&A across diverse topics

Limitations

Knowledge cutoff is based on training data
Not suitable for production-critical applications without testing
May not match the performance of much larger models on highly specialized tasks

Citation

bibtex
@misc{ult2max2026,
  title={Ult2 Max: Efficient Expert AI on Consumer Hardware},
  author={teo},
  year={2026},
  howpublished={\url{https://huggingface.co/teolm30/Ult2-Max}},
}

Acknowledgments

Built on Qwen2.5-3B-Instruct by Alibaba Cloud
Inspired by research from Anthropic, OpenAI, Google DeepMind, and other frontier AI labs

License

Apache 2.0 - Free for commercial and personal use.

Ult2 Max - Maximum intelligence, minimum resources.

Available on FriendliAI

Dedicated Endpoints

Run this model inference on single tenant GPU with unmatched speed and reliability at scale.

Learn more

Container

Run this model inference with full control and performance in your environment.

Learn more

Model Details

Model Provider

teolm30

Model Tree

Base

Qwen/Qwen2.5-3B-Instruct

Quantized

this model

Input Modalities

Text

Output Modalities

Text

Supported Functionality

Dedicated EndpointsContainer

Explore FriendliAI today

Get started Talk to an engineer

README

License: apache-2.0

Overview

Key Features

Expert-Level Intelligence: Trained on advanced reasoning, coding, math, science, and general knowledge
Ultra-Efficient: Only ~3B parameters - runs on CPUs, laptops, and basic hardware
Hardcoded System Prompt: Expert persona is built directly into the model configuration
Versatile: Handles programming, analysis, creative writing, math, and more
Safe & Responsible: Built with safety guidelines and responsible AI practices

Model Details

Table with columns: Property, Value
Property	Value
Base Model	Qwen/Qwen2.5-3B-Instruct
Parameters	~3.09B
Architecture	Qwen2ForCausalLM
Context Length	32,768 tokens
Language	English
License	Apache 2.0

GGUF Downloads (Recommended for Local Use)

GGUF files are optimized for local inference with llama.cpp, Ollama, and other compatible runtimes. This is the recommended format for running Ult2 Max on your own hardware.

Table with columns: Quantization, Size, Description
Quantization	Size	Description
Q4_K_M	~1.83 GB	Recommended - Best balance of quality and size. Fits in ~2GB RAM.
F16	~6.18 GB	Full precision - Maximum quality, requires ~8GB RAM.

Running with llama.cpp

bash
# Clone and build llama.cpp (or download pre-built binary)
git clone https://github.com/ggerganov/llama.cpp
cd llama.cpp && make

# Download the Q4_K_M GGUF
curl -LO https://huggingface.co/teolm30/Ult2-Max/resolve/main/Ult2-Max-Q4_K_M.gguf

# Run inference
./llama-cli -m Ult2-Max-Q4_K_M.gguf -p "What is quantum computing?" -n 256

Running with Ollama

Download the Q4_K_M GGUF:

bash
curl -LO https://huggingface.co/teolm30/Ult2-Max/resolve/main/Ult2-Max-Q4_K_M.gguf

Create a Modelfile:

markdown
FROM ./Ult2-Max-Q4_K_M.gguf
SYSTEM "You are Ult2 Max, an expert AI assistant."

Create and run:

bash
ollama create ult2max -f Modelfile
ollama run ult2max

Running with Python (Transformers)

python
from transformers import AutoModelForCausalLM, AutoTokenizer

model = AutoModelForCausalLM.from_pretrained(
    "teolm30/Ult2-Max",
    torch_dtype="auto",
    device_map="auto"
)
tokenizer = AutoTokenizer.from_pretrained("teolm30/Ult2-Max")

messages = [
    {"role": "user", "content": "Explain quantum computing in simple terms."}
]

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

outputs = model.generate(**inputs, max_new_tokens=512)
response = tokenizer.decode(outputs[0], skip_special_tokens=True)
print(response)

Hardware Requirements

Table with columns: Format, Minimum RAM, Recommended RAM
Format	Minimum RAM	Recommended RAM
Q4_K_M GGUF (CPU)	2 GB	4 GB
Q4_K_M GGUF (GPU)	2 GB VRAM	4 GB VRAM
F16 GGUF	8 GB	16 GB
SafeTensors (Transformers)	4 GB	8 GB GPU

Training & Development

Ult2 Max is built through a combination of techniques:

Base Model: Start with Qwen2.5-3B-Instruct, a highly capable small language model
System Prompt Integration: Expert-level system prompt hardcoded into the model configuration
Knowledge Distillation: Training data derived from analysis of advanced AI systems
Optimization: Model optimized for efficiency while maintaining quality

Capabilities

Programming: Write, debug, and explain code in multiple languages
Mathematics: Solve equations, explain concepts, step-by-step reasoning
Science: Explain complex scientific concepts clearly
Creative Writing: Stories, poetry, scripts, and more
Analysis: Data analysis, research, and problem-solving
General Knowledge: Factual Q&A across diverse topics

Limitations

Knowledge cutoff is based on training data
Not suitable for production-critical applications without testing
May not match the performance of much larger models on highly specialized tasks

Citation

bibtex
@misc{ult2max2026,
  title={Ult2 Max: Efficient Expert AI on Consumer Hardware},
  author={teo},
  year={2026},
  howpublished={\url{https://huggingface.co/teolm30/Ult2-Max}},
}

Acknowledgments

Built on Qwen2.5-3B-Instruct by Alibaba Cloud
Inspired by research from Anthropic, OpenAI, Google DeepMind, and other frontier AI labs

License

Apache 2.0 - Free for commercial and personal use.

Ult2 Max - Maximum intelligence, minimum resources.