Vikhr-Llama-3.2-1B-Instruct GGUF
GGUF quantizations of Vikhrmodels/Vikhr-Llama-3.2-1B-Instruct.
Quantization Table
| File | Bits | Size | Use Case |
|---|---|---|---|
model_f16.gguf | 16 | ~2.5 GB | Maximum quality, reference |
model_q8_0.gguf | 8 | ~1.3 GB | Best quality, near-lossless |
model_q6_k.gguf | 6 | ~1.0 GB | Great quality |
model_q5_k_m.gguf | 5 | ~0.9 GB | Balanced quality/size |
model_q5_k_s.gguf | 5 | ~0.85 GB | Slightly smaller Q5 |
model_q4_k_m.gguf | 4 | ~0.77 GB | Recommended, good balance |
model_q4_k_s.gguf | 4 | ~0.72 GB | Smaller Q4 |
model_q3_k_l.gguf | 3 | ~0.65 GB | Low VRAM, decent quality |
model_q3_k_m.gguf | 3 | ~0.60 GB | Low VRAM |
model_q3_k_s.gguf | 3 | ~0.55 GB | Very low VRAM |
model_q2_k.gguf | 2 | ~0.45 GB | Minimum size, lowest quality |
VRAM Requirements
| Quant | VRAM |
|---|---|
| F16 | ~3 GB |
| Q8_0 | ~1.8 GB |
| Q6_K | ~1.5 GB |
| Q5_K_M | ~1.3 GB |
| Q4_K_M | ~1.1 GB |
| Q3_K_M | ~0.8 GB |
| Q2_K | ~0.6 GB |
Usage
llama.cpp
./llama-cli -m model_q4_k_m.gguf -p "Your prompt here" -n 512
llama-cpp-python
from llama_cpp import Llama
llm = Llama(
model_path="model_q4_k_m.gguf",
n_ctx=2048,
n_gpu_layers=-1
)
response = llm(
"Your prompt here",
max_tokens=512,
stop=["<|eot_id|>"]
)
print(response["choices"][0]["text"])
LM Studio
Download any .gguf file and load it directly in LM Studio.
Ollama
ollama run hf.co/tinyopsec/Vikhr-Llama-3.2-1B-Instruct-GGUF
Original Model
Vikhrmodels/Vikhr-Llama-3.2-1B-Instruct
Quantized using llama.cpp.