howtospark
models / qwen3-6-35b-a3b-heretic

Qwen3.6 35B-A3B Heretic (uncensored)

tvall43/Qwen3.6-35B-A3B-heretic
params
35.1B
active params
3.0B
sparsity
MoE
attention
Full attention
architecture
qwen3_5_moe
license
apache-2.0
downloads
366
likes
22
hf updated
Apr 16, 2026

Visualize

total size

each column is a layer — lit blocks are the experts that fire per token

bits / param
16-bit
footprint
65.4 GiB
sparks to fit
block ≈
274M params
active paramsdormant expertsghost = precision traded awayblock fill ∝ bits kept · Qwen3.6 35B-A3B Heretic (uncensored)35.107181936B params