vamazing/Koa-AI-v1-Code-3B

Koa-AI v1 Code 3B

Koa-AI-v1-Code-3B is a fine-tuned 3B parameter language model built on the Ministral 3B architecture. It is optimized for lightweight text generation and coding tasks.

---

Model Details

• Developed by: vamazing
• Model Type: Causal Language Model / Conditional Generation
• Architecture: Ministral 3 (3B parameters)
• Language(s): English
• License: Apache 2.0
• Finetuned from: mistralai/Ministral-3b-instruct

---

Training Parameters

Feature	Specification
Base Architecture	Ministral 3B Instruct 2512
Parameters	~3 Billion
Precision / Format	4-bit NormalFloat (NF4) / BF16
Native Context Length	Up to 128,000 tokens (Fine-tuned at 2,048 sequence cap)
Fine-Tuning Method	QLoRA 4-bit (Quantized Low-Rank Adaptation)
Target Modules	q_proj, k_proj, v_proj, o_proj, gate_proj, up_proj, down_proj
LoRA Parameters	r = 16, alpha = 32, Dropout = 0.0
Primary Frameworks	Unsloth, PyTorch, Hugging Face Transformers
Learning Rate	0.0002
Optimizer	AdamW 8-bit

---

Quickstart & Usage

1. Running with transformers (Python)

To run the model using Hugging Face transformers (v5.5.0 or compatible):

import torch
from transformers import AutoTokenizer, AutoModelForCausalLM

model_id = "vamazing/Koa-AI-v1-Code-3B"

tokenizer = AutoTokenizer.from_pretrained(model_id, trust_remote_code=True)
model = AutoModelForCausalLM.from_pretrained(
    model_id,
    torch_dtype=torch.float16,
    device_map="auto",
    trust_remote_code=True
)

prompt = "Write a Python function to check if a number is prime."
messages = [{"role": "user", "content": prompt}]
formatted_prompt = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)

inputs = tokenizer(formatted_prompt, return_tensors="pt").to(model.device)
outputs = model.generate(**inputs, max_new_tokens=256, do_sample=True, temperature=0.7)

print(tokenizer.decode(outputs[0], skip_special_tokens=True))