Muse-Glimmer-30B - DASH-Q 2-bit GGUF
llama.cpp GGUF files of meta-models/Muse-Glimmer-30B 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 |
|---|---|---|---|
Muse-Glimmer-30B-DASHQ-IQ2_XXS.gguf |
IQ2_XXS | 8.16 GB | 2.34 |
Muse-Glimmer-30B-DASHQ-IQ2_XS.gguf |
IQ2_XS | 9.21 GB | 2.64 |
Muse-Glimmer-30B-DASHQ-IQ2_M.gguf |
IQ2_M | 10.03 GB | 2.88 |
Muse-Glimmer-30B-DASHQ-Q2_K_XL.gguf |
Q2_K_XL | 10.91 GB | 3.13 |
Text only: the vision tower is not included.
Perplexity (lower is better)
| Type | Model | Size | WikiText-2 | C4 |
|---|---|---|---|---|
| Q8_0 | 29.6 GB | 5.64 | 10.06 | |
| IQ2_XXS | llama.cpp IQ2_XXS (imatrix) | 7.96 GB | 8.12 | 15.38 |
| IQ2_XXS | unsloth UD-IQ2_XXS | 10.75 GB | 6.29 | 11.37 |
| IQ2_XXS | DASH-Q IQ2_XXS | 8.16 GB | 7.06 | 12.73 |
| IQ2_XS | llama.cpp IQ2_XS (imatrix) | 8.71 GB | 7.11 | 13.40 |
| IQ2_XS | unsloth UD-IQ2_XS | 11.51 GB | 6.09 | 10.98 |
| IQ2_XS | DASH-Q IQ2_XS | 9.21 GB | 6.41 | 11.51 |
| IQ2_M | llama.cpp IQ2_M (imatrix) | 9.85 GB | 6.43 | 11.71 |
| IQ2_M | unsloth UD-IQ2_M | 12.26 GB | 5.93 | 10.64 |
| IQ2_M | DASH-Q IQ2_M | 10.03 GB | 6.16 | 11.08 |
| Q2_K_XL | llama.cpp Q2_K (imatrix) | 10.69 GB | 6.33 | 11.39 |
| Q2_K_XL | unsloth UD-Q2_K_XL | 12.44 GB | 6.04 | 10.83 |
| Q2_K_XL | DASH-Q Q2_K_XL | 10.91 GB | 6.11 | 10.97 |
llama-perplexity, context 2048; WikiText-2 test, C4 validation (256 x 2048 tokens).
Zero-shot accuracy (lm-eval-harness, 0-shot)
| File | arc_easy | arc_challenge | piqa | hellaswag | winogrande | openbookqa | commonsense_qa | truthfulqa_mc2 | lambada_openai | avg |
|---|---|---|---|---|---|---|---|---|---|---|
| DASH-Q Q2_K_XL | 83.2 | 57.4 | 82.3 | 80.7 | 77.7 | 44.6 | 72.6 | 56.7 | 73.9 | 69.91 |
Usage
llama-cli -m Muse-Glimmer-30B-DASHQ-IQ2_M.gguf -ngl 99 -c 8192
License
Inherits the license of the base model (meta-models/Muse-Glimmer-30B).