Modelpublic

tinyopsec/MiniCPM5-2B-SFT-Pashto-GGUF

sourceHugging Faceapache-2.0updated 12d agoView on Hugging Face

1likes1.2kdownloads

Model Card

MiniCPM5-2B-SFT-Pashto-GGUF

GGUF quantizations of nassimjp/MiniCPM5-2B-SFT-Pashto, a fine-tuned version of openbmb/MiniCPM5-2B for Pashto language.

Quant Table

File	Bits	Est. Size	Use Case
model_f16.gguf	16	~4.6 GB	Reference / re-quantization
modelq80.gguf	8	~2.4 GB	Max quality, enough RAM
modelq6k.gguf	6	~1.9 GB	High quality
modelq5k_m.gguf	5	~1.6 GB	Balanced
modelq5k_s.gguf	5	~1.6 GB	Balanced, smaller
modelq4k_m.gguf	4	~1.4 GB	Recommended
modelq4k_s.gguf	4	~1.3 GB	Lower RAM
modelq3k_l.gguf	3	~1.1 GB	Low RAM
modelq3k_m.gguf	3	~1.0 GB	Low RAM
modelq3k_s.gguf	3	~0.9 GB	Minimum quality
modelq2k.gguf	2	~0.7 GB	Very low RAM only

VRAM / RAM Requirements

Quant	RAM
Q8_0	~3 GB
Q4KM	~2 GB
Q2_K	~1.5 GB

Usage

llama.cpp

bash

./llama-cli -m model_q4_k_m.gguf -p "Your prompt here" -n 256

llama-cpp-python

python

from llama_cpp import Llama
llm = Llama(model_path="model_q4_k_m.gguf")
output = llm("Your prompt here", max_tokens=256)
print(output["choices"][0]["text"])

LM Studio

Download any .gguf file and load directly in LM Studio.

Ollama

bash

ollama run hf.co/tinyopsec/MiniCPM5-2B-SFT-Pashto-GGUF:Q4_K_M

Notes

—Architecture: LlamaForCausalLM
—Fine-tuned for Pashto (پښتو) language
—Based on MiniCPM5-2B with hybrid reasoning (Think / No-Think modes)
—For llama.cpp, recommended: --min-p 0.0 to avoid repetition