Quanter's Note: Sibling of
sandeshrajx/Qwen3.5-24B-A3B-REAP-0.32, this one has been passed over completely for reasoning distills, I'd be interested to see what difference 7b more active experts can make.
GGUF quantization of
sandeshrajx/qwen3.5b-24b-a10b — a 24B-parameter MoE model with 10B active parameters per token. Architecture is Qwen3.5 MoE.
Source: sandeshrajx/qwen3.5b-24b-a10b
Converted by: QuantZhai benchmark pipeline
Quantization: IQ4_NL (importance-matrix 4-bit non-linear)
Hardware: dual-GPU (RTX 3080 10GB + V100-SXM2 32GB, 42 GB total)
Engine: llama.cpp with TurboQuant KV (q8_0 K / turbo3 V)
Perplexity:
macvox68 code corpus, ctx=4096, stride=512
1llama-cli -m qwen3.5b-24b-a10b-IQ4_NL.gguf \
2 -p "Write a mergesort in Python" \
3 -n 1024 -t 12 --temp 0.6 --top-p 0.95
4
5llama-server -m qwen3.5b-24b-a10b-IQ4_NL.gguf \
6 --host 0.0.0.0 --port 8080 -ngl 99 -t 12 \
7 --cache-type-k q8_0 --cache-type-v turbo3
Recommended: temp 0.6, top-p 0.95, context up to 256K.
MIT (this quantization).
Source model by
sandeshrajx — review its license separately.
