ICONN e1: The new era of Open-Source AI
GPU poor? Less than 3x A100s? A e1 Lite model is coming with just 22B parameters alongside a model for consumer CPUs with 14B and 7B parameters.
-
Emotional Context Awareness
ICONN e1 interprets emotional cues and adjusts tone, vocabulary, and response style—offering a more human-like, emotionally reactive experience. -
ICONN Emotional Core (IEC) (Notice: Not available on Huggingface)
Powered by millions of small AI agents, IEC gives ICONN its emotional personality, with billions of simulated emotional states and detections. -
Reasoning
ICONN e1 is one of the most powerful reasoning open-source models, and most closed-source models in or out of Huggingface.
What is in the ICONN i1 MoE?
ICONN i1 MoE and Experts
ICONN e1, being a MoE just like it's base model ICONN 1, has multiple expert models. Keywords are taken from the user's input to choose which expert generates the output.
| Expert Chosen | User Input |
|---|---|
| ICONN-e1 | 'Hi!' |
| ICONN-e1-Pro | Solve for m: m² − (2 + ∑₍ⱼ₌₁₎² j)·m + (1 + ∑₍ⱼ₌₁₎³ j² − 14) = 0. |
| ICONN-e1-Science | If a stable isotope of Ununoctium (Uuo, now Og) could be synthesized in bulk, what would be its most likely physical state at STP and why, considering relativistic effects? |
| ICONN-e1-Code | Create a zero-dependency quantum-safe VM in Zig that compiles a domain-specific language into a fully homomorphic encrypted IR, supports hot-reloading WebAssembly modules, parallel scheduling via lock-free fibers, and performs live introspection through a headless OpenGL debug overlay. |
ICONN-e1:
ICONN's general-purpose reasoning model, designed for everyday tasks, logic, and conversation.
ICONN-e1-Pro:
ICONN's advanced reasoning model, optimized for complex problem-solving in math, logic, and professional domains.
ICONN-e1-Science:
ICONN's scientific expert model, trained on advanced science datasets to enhance precision in physics, chemistry, biology, and technical reasoning.
ICONN-e1-Code:
ICONN's coding specialist, trained for programming, compiler theory, software architecture, and technical code generation across multiple languages.
Usage
**First, make sure you have at least 4x Nvidia A100 or a single B100, and 120GB RAM and 120-192GB VRAM. Don't have this? Use our Lite model, coming soon.
Run the code below to run ICONN i1:
from transformers import AutoTokenizer, AutoModelForCausalLM, pipeline
import torch
def run_iconn_chatbot(model_name="ICONNAI/ICONN-e1"):
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForCausalLM.from_pretrained(model_name)
device = 0 if torch.cuda.is_available() else -1
chat_pipeline = pipeline(
"text-generation",
model=model,
tokenizer=tokenizer,
device=device,
max_length=1624,
do_sample=True,
top_p=0.9,
temperature=0.4,
pad_token_id=tokenizer.eos_token_id
)
print(f"ICONN chatbot running with model: {model_name}. Type 'exit' to quit.")
conversation_history = ""
while True:
user_input = input("You: ")
if user_input.lower() == "exit":
print("Goodbye!")
break
conversation_history += f"User: {user_input}\nBot:"
response = chat_pipeline(conversation_history, max_length=len(tokenizer.encode(conversation_history)) + 100)[0]['generated_text']
bot_reply = response[len(conversation_history):].strip().split("\n")[0]
print(f"Bot: {bot_reply}")
conversation_history += f" {bot_reply}\n"
if __name__ == "__main__":
run_iconn_chatbot()
Cite Us
If you use ICONN 1, please cite us as follows:
@misc{iconnai_2025,
author = { ICONNAI },
title = { ICONN-e1-Beta (Revision ca41146) },
year = 2025,
url = { https://huggingface.co/ICONNAI/ICONN-e1-Beta },
doi = { 10.57967/hf/5861 },
publisher = { Hugging Face }
}