L1-Qwen3-8B-Exact GGUF
GGUF quantizations of l3lab/L1-Qwen3-8B-Exact.
Quantization Variants
| File | Bits | Size | Use Case |
|---|---|---|---|
model_f16.gguf |
16 | ~16 GB | Reference, maximum quality |
model_q8_0.gguf |
8 | ~8.5 GB | High quality, still large |
model_q6_k.gguf |
6 | ~6.5 GB | Balanced quality/size |
model_q5_k_m.gguf |
5 | ~5.3 GB | Good quality, recommended |
model_q5_k_s.gguf |
5 | ~5.1 GB | Slightly faster Q5 |
model_q4_k_m.gguf |
4 | ~4.3 GB | Standard quantization |
model_q4_k_s.gguf |
4 | ~4.1 GB | Slightly faster Q4 |
model_q3_k_l.gguf |
3 | ~3.3 GB | Small model variant |
model_q3_k_m.gguf |
3 | ~3.2 GB | Balanced 3-bit |
model_q3_k_s.gguf |
3 | ~3.0 GB | Fast 3-bit |
model_q2_k.gguf |
2 | ~2.3 GB | Minimal size, CPU-friendly |
VRAM Requirements
| Quantization | GPU VRAM | RAM (CPU) |
|---|---|---|
| F16 | ~16 GB | ~20 GB |
| Q8_0 | ~9 GB | ~11 GB |
| Q6_K | ~7 GB | ~9 GB |
| Q5_K_M | ~6 GB | ~8 GB |
| Q4_K_M | ~5 GB | ~6 GB |
| Q3_K_M | ~3.5 GB | ~5 GB |
| Q2_K | ~2.5 GB | ~3.5 GB |
Usage
llama.cpp
./main -m model_q5_k_m.gguf -n 256 -p "Your prompt here"
llama-cpp-python
from llama_cpp import Llama
llm = Llama(
model_path="model_q5_k_m.gguf",
n_gpu_layers=-1, # Use GPU
n_ctx=2048,
)
response = llm("Your prompt here", max_tokens=256)
print(response["choices"][0]["text"])
LM Studio
- Download any
.gguffile from this repo - Open in LM Studio
- Configure context and parameters
- Start chat
Ollama
ollama create l1-qwen3-8b -f Modelfile
Modelfile:
FROM model_q5_k_m.gguf
PARAMETER temperature 0.7
Base Model
l3lab/L1-Qwen3-8B-Exact — Fine-tuned reasoning model based on Qwen3 8B, trained with reinforcement learning to optimize reasoning step count.
License
Apache 2.0 (as per base model)