Downloads · 30 days
21
9% of all-time downloads
OmAlve/vaarta-gguf
vaarta-gguf is a machine learning model from OmAlve. Use it for the machine learning task on the model card, and read the license before you ship it in a product. The card lists the license as llama3.2.
Quantized GGUF files for OmAlve/vaarta-llama-instruct, a Llama-3.2-3B model fine-tuned on Marathi instruction data. Optimized for on-device inference with llama.cpp.
Downloads · 30 days
21
9% of all-time downloads
All-time downloads
246
Public
Repo size
4.3 GB
Likes
0
Public
Click a slice to open those files.
.gguf4.3 GB · 100%
From the Hugging Face model README
Quantized GGUF files for OmAlve/vaarta-llama-instruct, a Llama-3.2-3B model fine-tuned on Marathi instruction data. Optimized for on-device inference with llama.cpp.
| File | Quant | Size | Recommended for |
|---|---|---|---|
vaarta-Q4_K_M.gguf | Q4_K_M | 1.9 GB | ✅ Most Android phones (4 GB+ RAM) |
vaarta-Q5_K_M.gguf | Q5_K_M | 2.3 GB | Better quality (6 GB+ RAM) |
This model uses a custom Marathi Alpaca template — NOT the standard Llama-3 chat format.
खाली एक सूचना दिली आहे. विनंती पूर्ण करणारे योग्य उत्तर लिहा.
### सूचना:
{instruction}
### इनपुट:
{input}
### उत्तर:
{output}
The template is embedded in the GGUF metadata.
pkg update && pkg upgrade -y
pkg install -y clang cmake git wget
git clone https://github.com/ggerganov/llama.cpp
cd llama.cpp
cmake -B build -DGGML_NATIVE=OFF -DCMAKE_BUILD_TYPE=Release
cmake --build build --target llama-cli -j4
cd ~
wget https://huggingface.co/OmAlve/vaarta-gguf/resolve/main/vaarta-Q4_K_M.gguf
~/llama.cpp/build/bin/llama-cli \
-m vaarta-Q4_K_M.gguf \
--ctx-size 2048 --threads 4 \
-i \
-r "### सूचना:" \
--in-prefix "खाली एक सूचना दिली आहे. विनंती पूर्ण करणारे योग्य उत्तर लिहा.\n\n### सूचना:\n" \
--in-suffix "\n\n### इनपुट:\n\n\n### उत्तर:\n"
Type your Marathi question and press Enter.
./llama-cli -m vaarta-Q4_K_M.gguf -p "खाली एक सूचना दिली आहे. विनंती पूर्ण करणारे योग्य उत्तर लिहा.
### सूचना:
पृथ्वीबद्दल थोडक्यात माहिती द्या.
### इनपुट:
### उत्तर:
" -n 256