You can deploy this model and start earning money today!

KeefeBuild/
Keefe-Discere

Name: Keefe-Discere
Rating: 2.5 (1724 reviews)

HuggingFaceH4/ultrachat_200k

openai/gsm8k

tatsu-lab/alpaca

Views

No views yet

Keefe-Discere

An 8B-class instruction-following language model focused on reasoning, coding, mathematics, and agentic tool use.

Quick Start • Model Info • Training Data • Roadmap • Citation

Overview

Keefe-Discere is an independently developed language-model project by KeefeBuild, built on the Qwen2.5-7B instruction-tuned architecture and enhanced through a two-stage process:

Model Merging (DARE-TIES): Combining general, coding, and mathematics specialists into a single balanced 15GB checkpoint.
Targeted Post-Training (QLoRA v1.1): Fine-tuning a LoRA adapter on curated instruction, reasoning, and code-execution data to improve agentic tool use and mathematical reliability.

The project is designed as a general-purpose, locally-deployable language model with an emphasis on:

🧠 Reasoning and structured problem solving
🔢 Mathematics and quantitative tasks
💻 Programming, debugging, and code execution
🛠️ Agentic tool use (Python execution, function calling)
💬 General instruction following
🏠 Private, self-hosted, and offline inference

Important: Keefe-Discere is an independent model project and is not an official Qwen model.

Quick Start

🤗 Transformers (Python)

python

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

# 1. Load the merged base model
base_model_id = "KeefeBuild/Keefe-Discere"
tokenizer = AutoTokenizer.from_pretrained(base_model_id)
base_model = AutoModelForCausalLM.from_pretrained(
    base_model_id, torch_dtype=torch.bfloat16, device_map="auto"
)

# 2. Attach the v1.1 LoRA adapter for enhanced coding/tool use
model = PeftModel.from_pretrained(base_model, base_model_id)

messages = [
    {"role": "system", "content": "You are Keefe-Discere. Write clean Python code and use print() to output final answers."},
    {"role": "user", "content": "Calculate the sum of the first 15 prime numbers."}
]

inputs = tokenizer.apply_chat_template(messages, add_generation_prompt=True, return_tensors="pt").to(model.device)
outputs = model.generate(inputs, max_new_tokens=512, temperature=0.1)
print(tokenizer.decode(outputs[0][inputs.shape[-1]:], skip_special_tokens=True))