Inkling 975B · Alis MLX Dynamic - a avlp12 Collection

avlp12 's Collections

GLM-5.3-Flash · Alis MLX

Qwen3.8-27B — MLX, multimodal preserved

Laguna S 2.1 · Alis MLX Dynamic

Motif-3-Beta · Alis MLX (Apple Silicon)

Inkling 975B · Alis MLX Dynamic

Qwen3.6 — Alis MLX Dynamic

Alis MLX — big models on Apple Silicon

Lance · Alis MLX

CyberRealistic Z-Image-Turbo v4 · mflux

Krea 2 Turbo · Alis MLX

Qwen3.5-35B-A3B · Alis builds

Kimi K2.7 Code · Alis MLX Dynamic

GLM-5.2 · Alis MLX Dynamic

Inkling 975B · Alis MLX Dynamic

updated Jul 24

Receipt-sealed layer-local ALIS-DWQ MLX quants of Inkling 975B multimodal MoE, built on 2x512GB M3 Ultra. Quality 3.7 / capacity 2.7 bpw / Q6 soon.

thinkingmachines/Inkling

Image-Text-to-Text • 952B • Updated Jul 23 • 250k • • 1.76k

Note Base model — 975B-class multimodal MoE (Apache-2.0), 66 hybrid layers, 256 routed + 2 shared experts, vision + audio towers.
avlp12/Inkling-975B-Alis-MLX-Dynamic-3.7bpw

Image-Text-to-Text • 947B • Updated Jul 21 • 122 • 2

Note Quality / golden-spot tier: 3.71 bpw, 409 GiB — experts 3-bit, entire non-expert skeleton exact BF16. Certified DWQ pass, 1 accepted layer improvement. Single 512 GB Mac.
avlp12/Inkling-975B-Alis-MLX-Dynamic-6.6bpw

Image-Text-to-Text • Updated Jul 21 • 23

Note Performance / maximum-fidelity tier: 6.60 bpw, 728 GiB two-box pipeline build — experts 6-bit alis-clip, everything else byte-exact BF16. Ships as stage0/stage1 artifacts for 2x512GB serving.
avlp12/Inkling-975B-Alis-MLX-Dynamic-2.7bpw

Image-Text-to-Text • 947B • Updated Jul 24 • 33

Note Capacity / size-optimal tier: 2.72 bpw, 299 GiB — 2-bit expert floor with targeted 3/4/6/8-bit promotions. Certified DWQ pass, 7 accepted layer improvements. Fits 320 GB+ Macs.