SmolLM3-3B - DASH-Q 2-bit GGUF
llama.cpp GGUF files of HuggingFaceTB/SmolLM3-3B quantized with DASH-Q, at several 2-bit-class sizes. Every file uses only standard llama.cpp tensor types (no tensor above 4 bits) and loads in any recent llama.cpp build.
| File | Type | Size | Bits / weight |
|---|---|---|---|
SmolLM3-3B-DASHQ-IQ2_XXS.gguf |
IQ2_XXS | 0.92 GB | 2.40 |
SmolLM3-3B-DASHQ-IQ2_XS.gguf |
IQ2_XS | 1.05 GB | 2.74 |
SmolLM3-3B-DASHQ-IQ2_M.gguf |
IQ2_M | 1.12 GB | 2.92 |
SmolLM3-3B-DASHQ-Q2_K_XL.gguf |
Q2_K_XL | 1.22 GB | 3.19 |
Perplexity (lower is better)
| Type | Model | Size | WikiText-2 | C4 |
|---|---|---|---|---|
| IQ2_XXS | llama.cpp IQ2_XXS (imatrix) | 0.93 GB | 16.47 | 26.79 |
| IQ2_XXS | unsloth UD-IQ2_XXS | 0.96 GB | 15.22 | 24.42 |
| IQ2_XXS | DASH-Q IQ2_XXS | 0.92 GB | 13.06 | 22.83 |
| IQ2_XS | llama.cpp IQ2_XS (imatrix) | 1.02 GB | 12.84 | 21.05 |
| IQ2_XS | DASH-Q IQ2_XS | 1.05 GB | 10.60 | 18.54 |
| IQ2_M | llama.cpp IQ2_M (imatrix) | 1.13 GB | 10.81 | 18.32 |
| IQ2_M | unsloth UD-IQ2_M | 1.15 GB | 10.31 | 17.47 |
| IQ2_M | DASH-Q IQ2_M | 1.12 GB | 9.93 | 17.46 |
| Q2_K_XL | llama.cpp Q2_K (imatrix) | 1.25 GB | 10.72 | 18.06 |
| Q2_K_XL | unsloth UD-Q2_K_XL | 1.29 GB | 9.83 | 16.69 |
| Q2_K_XL | DASH-Q Q2_K_XL | 1.22 GB | 9.62 | 16.96 |
llama-perplexity, context 2048; WikiText-2 test, C4 validation (256 x 2048 tokens).
Usage
llama-cli -m SmolLM3-3B-DASHQ-Q2_K_XL.gguf -ngl 99 -c 8192
License
Inherits the license of the base model (HuggingFaceTB/SmolLM3-3B).