Downloads · 30 days
284
46% of all-time downloads
khadim-hussain/qwen3-14b-stem-qa-gguf
qwen3-14b-stem-qa-gguf is a text generation model from khadim-hussain. Use it when you need the model to write or continue text. The card lists the license as apache-2.0.
GGUF quantized version of Qwen3-14B fine-tuned for STEM Q&A tasks. Ready for use with Ollama, llama.cpp, LM Studio, and other GGUF-compatible tools.
Downloads · 30 days
284
46% of all-time downloads
All-time downloads
613
Public
Repo size
38.5 GB
Likes
0
Public
Click a slice to open those files.
.gguf38.5 GB · 100%
From the Hugging Face model README
GGUF quantized version of Qwen3-14B fine-tuned for STEM Q&A tasks. Ready for use with Ollama, llama.cpp, LM Studio, and other GGUF-compatible tools.
Author: Khadim Hussain · Project write-up: STEM reasoning LLMs
| Model | Size | Description |
|---|---|---|
| qwen3-14b-stem-qa | ~514MB | LoRA adapter only |
| qwen3-14b-stem-qa-merged | ~28GB | Full merged model (bf16) |
| qwen3-14b-stem-qa-gguf | 9-30GB | GGUF quantized (this repo) |
| Filename | Quant | Size | Description |
|---|---|---|---|
qwen3-14b-stem-qa-f16.gguf | F16 | 29.5GB | Full 16-bit precision, best quality |
qwen3-14b-stem-qa-q4_k_m.gguf | Q4_K_M | 9GB | 4-bit quantization, good balance of size/quality |
Ollama 0.34.1 and later no longer load LoRA adapters, so this merged GGUF is the way to run the fine-tune in Ollama. What changed, and how to serve the adapter with llama.cpp instead: Running a LoRA adapter after Ollama removed support.
# Download the GGUF file and Modelfile
# Then create and run:
ollama create qwen3-14b-stem -f Modelfile
ollama run qwen3-14b-stem
FROM ./qwen3-14b-stem-qa-q4_k_m.gguf
SYSTEM """You are a helpful assistant specialized in STEM subjects."""
TEMPLATE """{{- if .System }}<|im_start|>system
{{ .System }}<|im_end|>
{{ end }}
{{- range .Messages }}
{{- if or (eq .Role "user") (eq .Role "assistant") }}<|im_start|>{{ .Role }}
{{ .Content }}<|im_end|>
{{ end }}
{{- end }}<|im_start|>assistant
"""
PARAMETER stop "<|im_end|>"
PARAMETER temperature 0.7
PARAMETER top_p 0.9
PARAMETER num_ctx 4096
PARAMETER num_predict 2048
The template loops over .Messages so multi-turn chat keeps its format. A template with only {{ .Prompt }} (this repo's first version) renders one question the same, but Ollama treats it as a legacy template and garbles the history from the second message on. Details: khadim.tech.
./llama-cli -m qwen3-14b-stem-qa-q4_k_m.gguf \
-p "<|im_start|>user\nWhat is DNA?<|im_end|>\n<|im_start|>assistant\n" \
-n 512
qwen3-14b-stem-qa-q4_k_m.gguf| Metric | Value |
|---|---|
| Base Model | Qwen/Qwen3-14B |
| Train Loss | 0.461 |
| Eval Loss | 0.692 |
| Training Examples | 4,260 |
| Evaluation Examples | 474 |
If you use this model, please cite:
@misc{hussain2026qwen3-stem,
author = {Hussain, Khadim},
title = {Qwen3-14B STEM Q&A: Fine-tuned for Science Education},
year = {2026},
publisher = {Hugging Face},
url = {https://huggingface.co/khadim-hussain/qwen3-14b-stem-qa-gguf}
}
Apache 2.0 (inherited from Qwen3)