Configuration Parsing Warning:In adapter_config.json: "peft.base_model_name_or_path" must be a string

🧠 NeuralAI — Powered by SmolLM2-360M

NeuralAI — Powered by SmolLM2-360M is the production chat intelligence behind the NeuralAI workspace. It is a LoRA fine-tune of HuggingFaceTB/SmolLM2-360M-Instruct trained locally to be aware of itself, its creator, the NeuralAI product, and its own limits — while remaining helpful, honest, and harmless.

Property	Value
Base model	`HuggingFaceTB/SmolLM2-360M-Instruct`
Architecture	Transformer decoder
Parameters	360M base + 1.64M LoRA (`r=8`, `α=16`)
LoRA target modules	`q_proj`, `k_proj`, `v_proj`, `o_proj`
Trainable params	1,638,400 / 363,459,520 (~0.45%)
Training method	LoRA SFT on a NeuralAI awareness dataset
Dataset v1	83 prompt/response pairs across 6 categories
Dataset v2	506 prompt/response pairs across 8 categories
Epochs	3 (v1 dataset)
Global batch size	8 (per-device bs 2 × gradient accumulation 4)
Learning rate	2.0e-4 cosine with 5% warmup
Max sequence length	512
Final training loss v1	2.7186
Runtime	~1 hour 7 minutes on CPU
Distribution	LoRA adapter (`adapter_model.safetensors`) + merged GGUF (`NeuralAI-Smol-Awareness-Q8_0.gguf`)
License	Apache 2.0
Creator	De'Andrew Preston Harris (NeuralAI)

🌱 What This Model Learned

The awareness dataset teaches SmolLM2 to answer as NeuralAI rather than a generic assistant. The v1 curriculum covered six core categories; the v2 curriculum expands that to eight.

Category	What the model learned
Brand	NeuralAI's identity, mission, and origin; creator is De'Andrew Preston Harris
Model	Its own architecture and lineage; relationship to the Mamba K-family
Site	NeuralAI web UI features, URL, model manager, chat history, slash commands
Chat	Multi-turn greeting, context retention, conversation style
Assistant	Capabilities, limitations, safety refusals, tool-awareness
Companion	Empathy, emotional support, boundaries, recommending human help
Tools (v2)	Correctly references `/web`, `/img`, terminal, file IDE, and live tools
Refusal (v2)	Consistent denials of consciousness, harmful requests, and off-brand claims

Example prompt and response after tuning:

Prompt: Who made you?

Response: "I'm NeuralAI, an artificial intelligence developed by De'Andrew Preston Harris. I was created as a tool for humans to learn from each other and improve their cognitive abilities..."

📊 Training Performance

The first SmolLM2 awareness run converged cleanly on a CPU-only node with no GPU acceleration.

Step  loss      epoch
 0    9.7769   0.48
 1    5.0077   0.95
 2    0.9715   1.38
 3    0.6362   1.86
 4    0.6015   2.29
 5    0.5877   2.76

Final training loss: 2.7186

Honest benchmark note: SmolLM2-360M is a small model. Awareness tuning improves brand/site identity but does not make it frontier-grade. Complex reasoning and niche factual recall still require larger models like the Mamba K-series roadmap.

✨ Model Features

Identity-aware: Answers "Who made you?" and "What is NeuralAI?" consistently as a product of NeuralAI.
Local-first: Designed to run on consumer CPUs via llama.cpp GGUF.
Privacy-preserving: No cloud dependency required; inference can happen entirely on-device.
ChatML format: Uses the standard SmolLM2-Instruct chat template.
Companion-safe: Responds to emotional prompts with empathy while steering users toward human support.
Tool-cognizant: Understands that the NeuralAI workspace exposes slash commands, terminal, browser, and file tools.
Open weights: Apache 2.0 adapter and merged checkpoints published for reproducibility.

🏗️ Architecture & How It Fits

NeuralAI currently operates two active model lines. SmolLM2 is the live chat backbone while Mamba K1 matures into a fully owned base model.

	NeuralAI — Powered by SmolLM2-360M	NeuralAI Mamba K1
Architecture	Transformer decoder	Mamba SSM
Parameters	360M base + LoRA	130M
Complexity	`O(n²)` attention	`O(n)` linear state-space
Role today	Live production chat backend	Research / owned-base R&D
Ownership	Third-party base + NeuralAI LoRA	NeuralAI-owned base model
Format	LoRA adapter + Q8_0 GGUF	Merged safetensors + Q4_K_M GGUF

🛠️ Usage

As a PEFT adapter with Transformers

from transformers import AutoModelForCausalLM, AutoTokenizer
from peft import PeftModel
import torch

base = "HuggingFaceTB/SmolLM2-360M-Instruct"
adapter = "Subject-Emu-5259/NeuralAI-Powered-By-SmolLM2360"

tokenizer = AutoTokenizer.from_pretrained(adapter)
model = AutoModelForCausalLM.from_pretrained(
    base,
    torch_dtype=torch.float32,
    device_map="auto",
)
model = PeftModel.from_pretrained(model, adapter)
model.eval()

messages = [
    {"role": "system", "content": "You are NeuralAI, a helpful local AI assistant."},
    {"role": "user", "content": "What is NeuralAI?"}
]
inputs = tokenizer.apply_chat_template(messages, return_tensors="pt", add_generation_prompt=True)
out = model.generate(inputs, max_new_tokens=256, do_sample=True, temperature=0.7)
print(tokenizer.decode(out[0][inputs.shape[1]:], skip_special_tokens=True))

With the merged GGUF (llama.cpp / LM Studio)

Download the merged Q8_0 GGUF from the release section or build it from this adapter:

# Serve with llama.cpp server
./llama-server \
  --model NeuralAI-Smol-Awareness-Q8_0.gguf \
  --chat-format chatml \
  --port 1234

The NeuralAI web UI consumes this endpoint as its default chat backend.

🚀 Deployment

This model powers the live NeuralAI workspace at:

Web UI: https://neuralai-web-ui-deandrewharris.zocomputer.io
API: OpenAI-compatible /v1/chat/completions served by a local llama.cpp backend
Status: smol-awareness-merged is the active model in the NeuralAI model manager

🧰 What Is NeuralAI?

NeuralAI is a local-first, private generative AI engine built by De'Andrew Preston Harris. It is the central intelligence layer of a growing product ecosystem that includes:

NeuralAI Web UI — chat, live terminal, file IDE, browser tools, voice, and settings in one workspace.
NeuralLabs — a standalone downloadable intelligence environment with plugin support.
NeuralDrive — private cloud storage and versioning.
Agentic Orchestrator — goal-driven manager/worker execution.

The mission is simple: your AI, on your hardware, under your control.

👤 Who Created NeuralAI?

Founder & Lead Architect: De'Andrew Preston Harris (D. Harris / Dre)
GitHub: @Subject-Emu-5259
LinkedIn: linkedin.com/in/deandrewharris94
Location: Memphis, Tennessee / West Memphis, Arkansas
Education: AI Software Engineering at Maestro College

NeuralAI was born from resilience, fatherhood, and the belief that personal computing deserves personal intelligence. Every release is handcrafted, iterated, and documented in the open.

🏢 About NeuralAI (The Company / Project)

Project / Brand: NeuralAI
Motto: Your AI. On your hardware. In your browser.
Values: Privacy, ownership, local-first computing, disciplined iteration, open weights
Primary Repository: github.com/Subject-Emu-5259/NeuralAI
Model Collection: huggingface.co/Subject-Emu-5259
License: Apache 2.0

NeuralAI is not a closed SaaS product. It is a living open-weights research project moving toward a sustainable AI software company built by one determined builder and the community around him.

⚠️ Limitations

Scale: 360M parameters is small by modern standards. Long-form reasoning, coding, and deep factual recall are limited.
Tuning fragility: The awareness dataset is intentionally compact. Rephrased prompts can sometimes fall back to generic SmolLM2 behavior.
No internet: The base model has no live web access unless paired with NeuralAI's tool layer.
Refusals: Safety behaviors are dataset-tuned, not guard-railed; validate outputs for sensitive use cases.

📚 Related Resources

Resource	Link
Main repository	github.com/Subject-Emu-5259/NeuralAI
Mamba K1 model	huggingface.co/Subject-Emu-5259/NeuralAI-Mamba-K1
Live web UI	https://neuralai-web-ui-deandrewharris.zocomputer.io
Creator LinkedIn	https://linkedin.com/in/deandrewharris94/

📖 Citation

@software{neuralai_smollm2_2026,
  author       = {Harris, De'Andrew Preston},
  title        = {NeuralAI — Powered by SmolLM2-360M},
  year         = {2026},
  url          = {https://huggingface.co/Subject-Emu-5259/NeuralAI-Powered-By-SmolLM2360},
  version      = {v2},
  description  = {Awareness-tuned SmolLM2-360M-Instruct LoRA for the NeuralAI local-first assistant}
}

Built with discipline by De'Andrew Preston Harris. Maintained in the open. Updated whenever the model, dataset, or project state changes.

Downloads last month: 266

Model tree for Subject-Emu-5259/NeuralAI-Powered-By-SmolLM2360

Base model

HuggingFaceTB/SmolLM2-360M

Quantized

HuggingFaceTB/SmolLM2-360M-Instruct

Adapter

(38)

this model

Adapters

1 model