adtc-agri-sft-qwen2.5-1.5b-GGUF

GGUF-quantized versions of EYEDOL/adtc-agri-sft-qwen2.5-1.5b for use with llama.cpp, Ollama, LM Studio, and other GGUF-compatible runtimes.

The source model is a QLoRA fine-tune of Qwen/Qwen2.5-1.5B for agricultural advisory Q&A, part of the ADTC 2026 project (Nigeria-focused smallholder farmer / extension officer use case). See the source model card for full training details — data sources, hyperparameters, and the training loss curve.

Files

Pick Q4_K_M for the smallest footprint, Q6_K for the closest fidelity to the original fp16 model, or Q5_K_M as a middle ground.

Usage

llama.cpp

./llama-cli -m model-Q4_K_M.gguf -p "Your farming question here" -n 200 -c 2048

(swap the filename for whichever quantization you downloaded)

Ollama

# Modelfile
FROM ./model-Q4_K_M.gguf

ollama create adtc-agri-qwen25 -f Modelfile
ollama run adtc-agri-qwen25 "Your farming question here"

Python (llama-cpp-python)

from llama_cpp import Llama
llm = Llama(model_path="model-Q4_K_M.gguf", n_ctx=2048)
out = llm("Your farming question here", max_tokens=200)
print(out["choices"][0]["text"])

Intended use & limitations

Same scope as the source model: an offline agricultural advisory baseline for smallholder farmers and extension officers in Nigeria. Not validated for general-purpose assistant use, and outputs should be treated as advisory rather than authoritative agronomic or financial guidance — always recommend in-person expert follow-up for high-stakes decisions.

Quantization trades a small amount of output quality for a smaller file size and faster CPU inference versus the unquantized model — see the table above for the size/quality trade-off across levels. If you need maximum fidelity, use EYEDOL/adtc-agri-sft-qwen2.5-1.5b directly instead.

---

Converted with llama.cpp's convert_hf_to_gguf.py + llama-quantize, generated 2026-09-22 07:51 UTC.