Kairos-V0.1-1.5B-Instruct

Introduction

Kairos is a fine-tuned AI assistant built by ACAI (Assisting with Compassion and Artificial Intelligence) -- the AI platform built exclusively for nonprofits and philanthropy.

Kairos is fine-tuned with nonprofit and philanthropy domain knowledge: fundraising, grants, donor stewardship, impact measurement, SDGs, program design, and operations. It identifies as Kairos and knows about the ACAI platform.

This is the first release in the Kairos model series.

Feature Details
Parameters 1.5B
Fine-Tuning Method QLoRA (4-bit, r=32, alpha=64)
Training Data 490+ curated examples including 42 long-form expert answers (identity, nonprofit domain, ACAI platform)
Context Length 2048 tokens
License Apache 2.0
Author ACAI
Platform buildacai.com

Quickstart

from transformers import AutoModelForCausalLM, AutoTokenizer

model_name = "buildacai/Kairos-V0.1-1.5B-Instruct"

tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForCausalLM.from_pretrained(
    model_name,
    torch_dtype="auto",
    device_map="auto",
)

SYSTEM_PROMPT = (
    "You are Kairos, an expert AI assistant created by ACAI "
    "(Assisting with Compassion and Artificial Intelligence) specializing in nonprofits, "
    "philanthropy, grants, fundraising, and NGO operations. "
    "Always respond in English only. "
    "Provide comprehensive, well-structured answers with numbered steps, bullet points, "
    "and practical examples. Include actionable recommendations and best practices. "
    "Never invent internal model codes, version numbers, or URLs that were not part of your training."
)

messages = [
    {"role": "system", "content": SYSTEM_PROMPT},
    {"role": "user", "content": "How should an NGO write a compelling grant proposal?"},
]

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=2048,
    temperature=0.72,
    top_p=0.9,
    top_k=40,
    repetition_penalty=1.12,
    do_sample=True,
)
response = tokenizer.batch_decode(outputs[:, inputs.input_ids.shape[-1]:], skip_special_tokens=True)[0]
print(response)

What Kairos Can Do

  • Answer questions about fundraising, grants, donor stewardship, and impact measurement
  • Help with grant proposals, budgets, logic models, and impact narratives
  • Explain nonprofit concepts (SDGs, IRIS+, M&E, governance, compliance)
  • Describe the ACAI platform and its features
  • Respond to general knowledge questions
  • Always identify as Kairos, created by ACAI

Training Details

Method

Kairos-V0.1 was fine-tuned using Unsloth + QLoRA on a single NVIDIA T4 GPU (Google Colab free tier).

  • LoRA rank (r): 32
  • LoRA alpha: 64
  • Target modules: q_proj, k_proj, v_proj, o_proj, gate_proj, up_proj, down_proj
  • Quantization: 4-bit NormalFloat (NF4)
  • Epochs: 8
  • Learning rate: 3e-4
  • Batch size: 2 (effective 8 with gradient accumulation)

Data

Fine-tuned on 490+ curated instruction-following examples including 42 long-form expert answers (200-400 words each with numbered steps, best practices, and practical recommendations). Training covers: Kairos identity reinforcement, ACAI platform details (buildacai.com), nonprofit/NGO/philanthropy domain (fundraising, grants, SDGs, IRIS+, stewardship, impact measurement, donor analytics, program design, governance), and conversational interactions.

About ACAI

ACAI stands for Assisting with Compassion and Artificial Intelligence. It is an AI platform built exclusively for nonprofits, NGOs, foundations, and philanthropy organizations. ACAI brings together chat, donor insights, strategy building, grant drafting, impact reports, task boards, and more in one workspace.

Learn more at buildacai.com.

Model Series

Model Status Description
Kairos-V0.1-1.5B-Instruct Released Test release, 1.5B parameters
Kairos-V1-7B-Instruct Planned Production release, 7B
Kairos-V1-70B-Instruct Planned Full production, 70B, NGO domain-specific

Use with Ollama

Convert to GGUF and run locally:

ollama create kairos -f Modelfile
ollama run kairos

License

This model is released under the Apache License 2.0.

Citation

@misc{kairos2026,
    title  = {Kairos-V0.1-1.5B-Instruct: A Fine-Tuned AI for Nonprofits},
    author = {ACAI},
    year   = {2026},
    url    = {https://huggingface.co/buildacai/Kairos-V0.1-1.5B-Instruct}
}

Contact

Downloads last month
45
Safetensors
Model size
2B params
Tensor type
BF16
·
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support

Model tree for buildacai/Kairos-V0.1-1.5B-Instruct

Quantizations
2 models