Skip to content

armgit / janas

Public Imported from GitHub
0

37 tok/s from Qwen3-Next-80B (48 GB) on a 32 GB laptop, 24 without drafts vs 2.8 for llama.cpp; reading 62 vs 16 tok/s. Faster than llama.cpp on all 13 models compared in Q4_K_M. LLMs bigger than your RAM, pure C, CPU or GPU.

NameLast commit date
..
attn_decode.comp
attn_merge.comp
b32_matvec.comp
b32_mmid.comp
bf16_matvec.comp
dense_mm.comp
dense_mv.comp
expert_act.comp
fma_probe.comp
gdn_act.comp
gdn_ba.comp
gdn_conv.comp
gdn_cstate.comp
gdn_gate.comp
gdn_l2.comp
gdn_rec.comp
hc_act.comp
hc_comb.comp
hc_mean.comp
hc_norm.comp
iq1_matvec.comp
iq2_matvec.comp
iq3s_matvec.comp
iq3s_mmq.comp
iq3xxs_matvec.comp
iq4nl_mmq.comp
iq4xs_matvec.comp
iq4xs_mmq.comp
iq4xs_mvd.comp
iqk_mmid.comp
iqk_mmq.comp
ix_pool.comp
ix_prep.comp
ix_score.comp
ix_select.comp
moe_combine.comp
moe_gather.comp
moe_plan.comp
moe_topk.comp
mxfp4_mmid.comp
ng_conv.comp
ng_gate.comp
norm_quant.comp
ple_act.comp
q2k_matvec.comp
q3k_matvec.comp
q3k_mmid.comp
q3k_mmq.comp
q4k_matvec.comp
q4k_mmid.comp
q4k_mmq.comp
q4k_mv.comp
q4k_mvd.comp
q4k_row.comp
q5k_matvec.comp
q5k_mmq.comp
q5k_mvd.comp
q6k_matvec.comp
q6k_mmid.comp
q6k_mmq.comp
q6k_mvd.comp
q6k_row.comp
q6ka_row.comp
q6kp_matvec.comp
q6kp_mmid.comp
q6kp_mmq.comp
q6kp_row.comp
q8_0_matvec.comp
q80_mmq.comp
qk_rope.comp
silu_mul.comp
xfma.glsl