Motif-3-Beta-Alis-MLX-8bit

Apple Silicon (MLX) 8-bit quantization of Motif-Technologies/Motif-3-Beta — a 314.84B-parameter (≈13B active) Mixture-of-Experts reasoning model, text-only, bilingual Korean / English (+ code), with 97.84% of its mass in the routed experts and native 256K context. This is the near-lossless reference build: uniform 8-bit (g64), 8.503 bpw measured, 312 GB on disk, for 512 GB Apple Silicon Macs. Part of the first community MLX port of Motif-3-Beta in any format.

This build is the anchor of the family: it is statistically near-lossless versus the bf16 original, and every other build's quality is measured against it (the bf16 model does not fit in memory, so Q8 is the practical ground truth). Run it when you want the maximum-quality MLX artifact or a KL/eval reference.

⚠️ Requires a patched mlx-lm with the motif model class (see How to run). Stock mlx-lm has no Motif support and will not load this checkpoint.

⚠️ Non-commercial research license, inherited in full from Motif-3-Beta (a beta release). See License.

---

Model at a glance

Motif-3-Beta is a text-generation LLM (no vision/audio) built on several non-standard components. The port reimplements each for MLX and verifies it against a bug-fixed reference.

Base model	Motif-Technologies/Motif-3-Beta — 314.84B total / ≈13B active
Architecture	Sparse MoE, 53 layers (0-1 dense, 2-52 MoE); 384 routed experts, top-8 + 1 shared
Routed-expert mass	97.84% of parameters
Attention	GDLA (Grouped Differential Latent Attention): MLA-style low-rank q/kv + differential-v2 + elementwise output gate; 80 heads / 16 KV groups, head_dim 192
Sliding window	interleaved SWA — window 128 on 40 layers; 13 full-attention layers carry YaRN, the 40 SWA layers plain rope
Other components	mHC (manifold-constrained hyper-connections, ×2/layer); Grouped PolyNorm per-expert activation; rms_norm eps 1e-6
Tokenizer	vocab 220,160; stop tokens {0, 3, 6}; chat template auto-opens <think>
Context	256K (262,144) native
Bits/weight	8.503 (measured; uniform 8-bit/g64)
On-disk size	312 GB (312 GiB)
Format	MLX (Apple Silicon); library_name: mlx
Role	near-lossless reference · KL / eval ground truth for the family

Recipe (verified from config.json)

Uniform 8-bit — no bit is starved. Even at 8-bit the discrete control paths are kept in full precision, as on every build in the family.

Component	Bits	Notes
Routed experts (gate/up/down)	8-bit g64	≈97.8% of params
Attention (GDLA projections)	8-bit g64	per-token critical path
wkv_a (KV-A latent down-projection)	8-bit g64	latent chokepoint
Token embedding · LM head	6-bit g64	distribution-sensitive
Router gate · mHC · lambda_proj · all norms · act_fn	fp (unquantized)	discrete control paths + normalization

Architecture

---

Build ladder

Build	Recipe	Size	bpw	Target
2.3bpw (FLOOR)	experts 2b/g128 · attn 4b · wkv_a 8b	85 GB	2.315	128 GB Mac
4.5bpw (C6)	experts 4b · attn 6b · wkv_a 8b	167 GB	4.547	256 / 512 GB Mac · DWQ teacher
8bit (this, Q8)	all 8b/g64 · embed/head 6b · router/mHC/norms fp	312 GB	8.503	512 GB Mac · reference

---

How to run

Motif support is not yet in an mlx-lm release. Install from the fork/branch that carries the motif model class:

pip install git+https://github.com/avlp12/mlx-lm.git@motif

Python:

from mlx_lm import load, generate
from mlx_lm.sample_utils import make_sampler

model, tokenizer = load("avlp12/Motif-3-Beta-Alis-MLX-8bit")

messages = [{"role": "user", "content": "한국어와 영어로 자기소개를 해줘."}]
prompt = tokenizer.apply_chat_template(messages, add_generation_prompt=True)
# The chat template auto-opens a <think> block: the model reasons first, then answers.
sampler = make_sampler(temp=1.0, top_p=0.95)   # recommended; greedy can loop on some prompts
print(generate(model, tokenizer, prompt=prompt, max_tokens=1024, sampler=sampler))

CLI / server:

mlx_lm.generate --model avlp12/Motif-3-Beta-Alis-MLX-8bit \
  --prompt "Explain MoE routing in two sentences." --temp 1.0 --top-p 0.95 -m 512

mlx_lm.server --model avlp12/Motif-3-Beta-Alis-MLX-8bit   # OpenAI-compatible

• Sampling: temperature 1.0, top-p 0.95 are recommended (greedy can loop on some prompts, even at 8-bit — it is a base-model tendency, not a quantization artifact).
• Stop tokens: the tokenizer stops on {0, 3, 6}; pass all three as stop ids if you drive generation yourself.
• Reasoning mode: the chat template opens a <think> block automatically — expected behavior, not a template bug.
• Long context: on 512 GB, quantize the KV cache for long prompts (--kv-bits 8 --kv-group-size 64 --quantized-kv-start 4096). See Hardware.

---

Method — alis-dwq pipeline

Weights are produced with alis-dwq, the same pipeline used for the GLM-5.2 (745B) and Hy3 (295B) Alis MLX builds. This build is a straight uniform 8-bit affine quantization (g64), with embed/head at 6-bit and the router / mHC / lambda_proj / norms kept fp — no clip-search or DWQ pass. It is the quasi-continuous source the more aggressive builds are quantized and distilled from, and the reference they are measured against.

The port itself is parity-verified: MLX logits match a bug-fixed HF reference to KL ≈1e-7 / token (see Port notes).

---

Evaluation

This build is the reference — near-lossless versus the bf16 original — so the family's KL and loop-probe numbers are measured against it, not for it. Fluent bilingual generation is confirmed: Korean (including hanja), English, and code, with the reasoning <think> trace intact. General knowledge, code, and bilingual generation are reliable.

The rest of the ladder, measured vs this 8bit reference (fixed EN / code / KO slice):

build	KL(Q8 ‖ build) EN / code / KO	notes
8bit (this)	reference (0 by definition)	near-lossless vs bf16
4.5bpw (C6)	teacher-grade — not separately measured	was the DWQ teacher; near-golden
2.3bpw (FLOOR)	0.277 / 0.100 / 0.522	after clip-search + DWQ

Everything in the family is measured against this build. The chart tracks the 2.3bpw floor build's recovery toward this reference through clip-search and DWQ.

Honest caveat (a base-model property, shown clearly by the reference): very-long-tail factual recall can be weak — e.g. the model does not reliably reproduce obscure text such as the 4th verse of the Korean national anthem. Because this build is near-lossless, that behavior is the base model's, not a quantization artifact; the smaller builds inherit it.

---

Correctness & port notes

• Logit parity. On a truncated model (fp32, EN/KO/code prompts) MLX matches a bug-fixed HF reference to KL ≈1e-7 / token, top-1 100%. SWA long-context self-consistency is a bit-exact match (cached wide forward vs no-cache reference).
• The shipped HF reference had bugs (documented in the repo discussion Motif-3-Beta/discussions/6): a YaRN RoPE dimension crash, an eager-attention GQA crash, and a grouped_mm path that applied expert-0 PolyNorm coefficients to all experts. Motif Technologies pushed fixes (SHA d2c9ac6) and this port matches the fixed reference. The root cause of the early incoherence was the Grouped PolyNorm activation (sigmoid coefficients + output_scale 0.5 + routed bias_clamp) — credit to Motif for the fix.
• One MLX-core bug is worked around in the port: mx.split silently corrupts the tail of a >2³¹-element tensor (the 8 GB bf16 gate_up); basic strided slices are used instead. Upstream ml-explore/mlx#3836.

---

Hardware

Built for 512 GB Apple Silicon (M3 Ultra class). Weights are 312 GB, leaving ample room for KV cache and long-context prefill activation within a 512 GiB machine; use int8 KV (--kv-bits 8) for the longest prompts. If your Mac has 256 GB, run the 4.5bpw build; for 128 GB, the 2.3bpw floor build.

---

License

Motif-3-Beta ships no formal license name and no LICENSE file; its model card states, verbatim:

"Permission is granted to use, modify, and redistribute this software for personal, educational, and non-commercial research purposes only. Commercial use is prohibited without prior written permission from Motif Technologies."

This is a non-commercial research license that explicitly permits modification and redistribution for personal / educational / non-commercial research use. A quantized MLX port is a modification-and-redistribution, so it is permitted provided it stays non-commercial and carries the same terms. This derivative inherits that license in full. The license: other / license_name fields are a best-effort mapping — the authoritative terms are the quoted text and the base model's card.

Credits & attribution

• Base model: Motif Technologies — Motif-3-Beta (non-commercial research). All architecture — GDLA, Grouped PolyNorm, mHC — and the trained weights are theirs, as are the reference fixes (SHA d2c9ac6).
• MLX & mlx-lm: Apple ml-explore.
• MLX port, motif model class, mixed-precision recipe: Alis (avlp12), via alis-dwq.

Independent, unofficial derivative — not affiliated with or endorsed by Motif Technologies. Provided as-is, without warranty. Research / non-commercial use only.

Citation

Alis (avlp12) (2026). Motif-3-Beta-Alis-MLX-8bit — near-lossless 8.503 bpw MLX reference build of Motif-3-Beta for 512 GB Apple Silicon; the family's KL / eval ground truth.