Instructions to use Shiftedx/qwen3.8-27b-aeon-ultimate-uncensored-attention8-bf16recurrence-vision-mtplx with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use Shiftedx/qwen3.8-27b-aeon-ultimate-uncensored-attention8-bf16recurrence-vision-mtplx with MLX:

# Make sure mlx-vlm is installed
# pip install --upgrade mlx-vlm

from mlx_vlm import load, generate
from mlx_vlm.prompt_utils import apply_chat_template
from mlx_vlm.utils import load_config

# Load the model
model, processor = load("Shiftedx/qwen3.8-27b-aeon-ultimate-uncensored-attention8-bf16recurrence-vision-mtplx")
config = load_config("Shiftedx/qwen3.8-27b-aeon-ultimate-uncensored-attention8-bf16recurrence-vision-mtplx")

# Prepare input
image = ["http://images.cocodataset.org/val2017/000000039769.jpg"]
prompt = "Describe this image."

# Apply chat template
formatted_prompt = apply_chat_template(
    processor, config, prompt, num_images=1
)

# Generate output
output = generate(model, processor, formatted_prompt, image)
print(output)

Notebooks
Google Colab
Kaggle
Local Apps Settings
LM Studio

How to use Shiftedx/qwen3.8-27b-aeon-ultimate-uncensored-attention8-bf16recurrence-vision-mtplx with Pi:

Start the MLX server

# Install MLX LM:
uv tool install mlx-lm
# Start a local OpenAI-compatible server:
mlx_lm.server --model "Shiftedx/qwen3.8-27b-aeon-ultimate-uncensored-attention8-bf16recurrence-vision-mtplx"

Configure the model in Pi

# Install Pi:
npm install -g @earendil-works/pi-coding-agent
# Add to ~/.pi/agent/models.json:
{
  "providers": {
    "mlx-lm": {
      "baseUrl": "http://localhost:8080/v1",
      "api": "openai-completions",
      "apiKey": "none",
      "models": [
        {
          "id": "Shiftedx/qwen3.8-27b-aeon-ultimate-uncensored-attention8-bf16recurrence-vision-mtplx"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Hermes Agent

How to use Shiftedx/qwen3.8-27b-aeon-ultimate-uncensored-attention8-bf16recurrence-vision-mtplx with Hermes Agent:

Start the MLX server

# Install MLX LM:
uv tool install mlx-lm
# Start a local OpenAI-compatible server:
mlx_lm.server --model "Shiftedx/qwen3.8-27b-aeon-ultimate-uncensored-attention8-bf16recurrence-vision-mtplx"

Configure Hermes

# Install Hermes:
curl -fsSL https://hermes-agent.nousresearch.com/install.sh | bash
hermes setup
# Point Hermes at the local server:
hermes config set model.provider custom
hermes config set model.base_url http://127.0.0.1:8080/v1
hermes config set model.default Shiftedx/qwen3.8-27b-aeon-ultimate-uncensored-attention8-bf16recurrence-vision-mtplx

Run Hermes

hermes

Atomic Chat

OpenClaw

How to use Shiftedx/qwen3.8-27b-aeon-ultimate-uncensored-attention8-bf16recurrence-vision-mtplx with OpenClaw:

Start the MLX server

# Install MLX LM:
uv tool install mlx-lm
# Start a local OpenAI-compatible server:
mlx_lm.server --model "Shiftedx/qwen3.8-27b-aeon-ultimate-uncensored-attention8-bf16recurrence-vision-mtplx"

Configure OpenClaw

# Install OpenClaw:
npm install -g openclaw@latest
# Register the local server and set it as the default model:
openclaw onboard --non-interactive --mode local \
  --auth-choice custom-api-key \
  --custom-base-url http://127.0.0.1:8080/v1 \
  --custom-model-id "Shiftedx/qwen3.8-27b-aeon-ultimate-uncensored-attention8-bf16recurrence-vision-mtplx" \
  --custom-provider-id mlx-lm \
  --custom-compatibility openai \
  --custom-text-input \
  --accept-risk \
  --skip-health

Run OpenClaw

openclaw agent --local --agent main --message "Hello from Hugging Face"

AEON Qwen3.8 27B — Attention8 + BF16 Recurrence for MLX/MTPLX

A quality-first Apple Silicon conversion of AEON-7/Qwen3.8-27B-AEON-ULTIMATE-UNCENSORED-BF16. It packages the mixed-precision language model, BF16 vision tower, tokenizer, tool template, and native BF16 multi-token-prediction head in one MTPLX-ready repository.

This is an independent community conversion, not an official AEON-7, Qwen, Apple, MLX, or MTPLX release.

At a glance

Property	Value
Model	27B dense hybrid-attention/Gated DeltaNet
Format	MLX safetensors: affine 4-bit + affine 8-bit + BF16
Download	23.84 GB (22.20 GiB)
Modalities	Text and images
Agent features	Reasoning, native tools, structured output, optional Shiftedx Agent Harness
Speculative decoding	Native 15-tensor BF16 MTP head; qualified at depth 3
Context	Configured for 262,144 tokens; qualified through 131,072 prompt tokens
Reference host	Apple M4 Max, 64 GiB unified memory
License	Apache-2.0, inherited from the source lineage

Use this release when you have a 64 GB-or-larger Apple Silicon Mac and want the quality-focused local build. If memory or storage is tighter, use the 16.08 GB MXFP4 sibling. The 29.53 GB MXFP8 sibling uses a larger, simpler precision strategy. The MXFP4+MTP sibling completed the same Shiftedx Bench v0.3.0 gate and tied this release across all five published lanes under its separately tuned D2 runtime contract. This Attention8 + BF16 recurrence artifact remains the quality-first precision recipe selected by the earlier parent-comparison suite; use the MXFP4 sibling when deployment size and memory matter more. Do not infer quality from file size alone.

Why this hybrid is larger than a 4-bit model

The quantization recipe spends precision where local regression tests found the architecture most sensitive:

168 language modules use affine 4-bit weights with group size 32.
234 attention, embedding, output, and late-MLP modules use affine 8-bit weights with group size 64.
96 recurrent in_proj_a and in_proj_b projections remain BF16.
The vision tower and matching 15-tensor MTP head remain BF16.

This is a custom MLX affine mixed-precision model—not MXFP4, MXFP8, NVFP4, GGUF, AWQ, or GPTQ. The exact module map is recorded in BUILD_RECIPE.json.

Qualification snapshot

The recipe-selection suite chose this hybrid over the compact local 4-bit baseline: hard-task accuracy improved from 6/10 to 8/10, repeated code checks remained 6/6, and code decode stayed close at 48.10 versus 50.56 tok/s. Full candidate comparisons are in BENCHMARKS.md.

MTPLX 2.9 performance retest

A performance-only retest of the exact published weight-bearing revision b5a54ea5d7745b6ddada238f83b66d63c979b9a5 used MTPLX 2.9.0 on the same Apple M4 Max 64 GiB host:

Lane	Result
Isolated D3 decode, three-run mean	63.36 tok/s
Isolated D3 decode, observed range	59.67–65.77 tok/s
Isolated D3 end-to-end, three-run mean	56.66 tok/s
OpenAI API code decode	54.83 tok/s
OpenAI API code end-to-end	45.83 tok/s
API code validation	3/3

The API code decode result is 14.0% above the original MTPLX 2.7.1 recipe-selection result of 48.10 tok/s. D3 remained the winner with 90.76% aggregate draft acceptance in the isolated lane. Testing used automatic fans; no privileged fan pinning or 80–90 tok/s claim is included. The isolated lane used the recommended temperature 1.0/top-p 0.95/top-k 20 target and draft samplers. The API code lane was a deterministic temperature-0, thinking-off diagnostic with unquantized KV cache.

This retest updates performance evidence only. The complete quality, context, tools, agentic, and vision qualification below remains the MTPLX 2.7.1 run. See BENCHMARKS.md and the MTPLX_2.9_PERFORMANCE.json sanitized aggregate.

MTPLX 2.9.2 MTP runtime retest

A second performance-only retest kept the exact model weights and native BF16 MTP sidecar unchanged, then quantized only the loaded MTP runtime block and draft-only vocabulary projection:

Configuration	Eight-sample median decode	Peak memory
BF16 MTP block + Q4/G64 draft head	38.70 tok/s	26.16 GB
Q6/G64 MTP block + Q4/G32 draft head	40.72 tok/s	25.73 GB

The selected D3 runtime improved median decode by 5.2% and reduced observed peak memory by approximately 0.43 GB. On a separate API hard-suite lane it improved mean decode by 3.7% and wall time by 2.8% with byte-identical responses under that lane's request contract. These are deployment-runtime results, not a weight-level quality improvement. The complete quality, context, tools, agentic, and vision score below remains bound to its original MTPLX 2.7.1 contract.

Post-publication Shiftedx Bench gate

Lane	Result
Quality	8/10
Long context	15/15, through 131,072 prompt tokens
Tool calling	6/6
Lightweight agentic	2/2
Vision	4/4

Evaluated model revision: c839d980679e9aec1ab66388ded0689040c9d93a
Benchmark revision: Shiftedx Bench v0.3.0 3bbb0bfa
Runtime: MTPLX 2.7.1 turbo, MTP depth 3, thinking on/medium, temperature 1.0, top-p 0.95, top-k 20, unquantized KV cache
Host: Apple M4 Max with 64 GiB unified memory

Scores are reported separately; Shiftedx Bench does not produce a composite “intelligence” score. The sanitized gate artifact is SHIFTEDX_BENCH_V0.3_TOOLS_AGENTIC_VISION.json.

Expanded agentic comparison

The two-case gate above is a regression check, not a broad agent-readiness claim. Shiftedx Bench v0.5.1 ran three complete paired trials with and without the optional shiftedx-harness-v1 execution policy:

Evaluation	Baseline	Shiftedx Agent Harness	Difference
30-case expanded matrix	54/90 (60.0%)	86/90 (95.6%)	+35.6 points
Disposable-repository repairs	9/12 (75.0%)	11/12 (91.7%)	+16.7 points

The harness converts observable tool receipts into bounded execution state. It does not change the weights or see hidden answers/tests. On the expanded matrix it reduced dispatched tool calls from 282 to 267 but increased total wall time by 11.4%. Repository trials were more variable; baseline won one of the three. Treat this as a higher-reliability deployment policy at a different operating point—not a weight-level intelligence improvement or a free speedup.

Evaluated model revision: b5a54ea5d7745b6ddada238f83b66d63c979b9a5
Benchmark revision: Shiftedx Bench v0.5.1 335e6694
Readable report, family scores, telemetry, sanitized per-case rows, and manifests

Quick start

Install MTPLX 2.9.2 or newer, then:

MODEL=Shiftedx/qwen3.8-27b-aeon-ultimate-uncensored-attention8-bf16recurrence-vision-mtplx

mtplx pull "$MODEL"
mtplx inspect "$MODEL" --require-mtp
mtplx start cli \
  --model "$MODEL" \
  --profile turbo \
  --depth 3 \
  --reasoning on \
  --reasoning-effort medium \
  --tool-prompt-mode native \
  --chat-template-profile tokenizer

For an OpenAI-compatible endpoint:

mtplx serve \
  --model "$MODEL" \
  --download \
  --host 127.0.0.1 \
  --port 8000 \
  --profile turbo \
  --generation-mode mtp \
  --load-mtp \
  --depth 3 \
  --context-window 198000 \
  --mtp-quant-bits 6 \
  --mtp-quant-group-size 64 \
  --mtp-quant-mode affine \
  --reasoning on \
  --reasoning-effort medium \
  --tool-prompt-mode native \
  --chat-template-profile tokenizer \
  --paged-kv-quantization off

The high-level mtplx serve wrapper reads the Q4/G32 draft-only projection from this artifact's recommended_draft_lm_head runtime metadata. Do not pass the lower-level --draft-lm-head-* server flags to the wrapper.

MTPLX accepts OpenAI tools and standard image_url content parts. For the published 6/6 tool result, ordinary/multi-round requests used the hybrid tool contract while explicit parallel intent used Qwen's native template. That router policy is included as MTPLX_2.7.1_QWEN38_EXPLICIT_PARALLEL_NATIVE.patch.

Recommended starting point: turbo profile, MTP depth 3, thinking on at medium effort, temperature 1.0, top-p 0.95, top-k 20, and unquantized KV cache. Run mtplx tune --model "$MODEL" --retune on your own Mac because acceptance and speed vary with chip, context, thermals, and workload.

The qualified default keeps thinking on at medium effort. For a low-latency request that must suppress reasoning, send the canonical top-level OpenAI field "enable_thinking": false. MTPLX 2.9.2 was verified to return zero reasoning tokens for that field while preserving ordinary stop completions, strict tool calls, and the long-context needle. Re-check per-request behavior after runtime upgrades. The separate thinking-off diagnostic scored 5/10 on its local ten-task hard suite; it does not replace the published 8/10 medium-thinking quality score.

Context and memory

The model advertises a 262,144-token context window and passed the exact-token qualification grid through 131,072 prompt tokens with unquantized KV cache. A 260,096-token attempt exceeded the practical memory envelope of the 64 GiB reference host and ended in a host watchdog. That is reported as not qualified on this host/runtime, not as a wrong model answer.

Reaching the full window may require a larger-memory Mac or a separately qualified q8/q4 KV-cache configuration. Changing KV precision creates a different runtime candidate and requires new quality, context, performance, and stability measurements. A 32 GiB Mac was not tested and may be too constrained; prefer the MXFP4 sibling when memory is tight.

Provenance and integrity

Source: AEON-7/Qwen3.8-27B-AEON-ULTIMATE-UNCENSORED-BF16
Pinned source revision: a6775a9a8ebb65cab3f707b4ab087fc7aa698634
BUILD_RECIPE.json: precision policy
mtplx_runtime.json: native-MTP contract and defaults
RELEASE_MANIFEST.json and SHA256SUMS: file integrity
AEON Qwen3.8 MLX collection

The vision tower and MTP sidecar are preserved from the qualified source lineage. Model behavior still depends on the exact runtime, template, sampler, reasoning mode, MTP depth, KV precision, and hardware.

Limitations and safety

Quantization can change outputs relative to BF16; qualification does not guarantee source parity on every prompt.
Tool calls, JSON, code, image interpretations, and factual answers can be wrong. Validate output before allowing system changes or code execution.
Only the Apple M4 Max 64 GiB configuration received the complete published qualification.
This intentionally uncensored source lineage may produce harmful, illegal, offensive, or unsafe material. Quantization does not restore alignment.

Operators are responsible for prompts, outputs, authentication, rate limits, moderation, audit logging, and human review appropriate to their deployment. Do not treat output as professional legal, medical, financial, or security advice.

License and credits

Apache-2.0, inherited from the source lineage. Follow the source model's license and attribution requirements when redistributing or deploying this derivative.

Credits to Qwen, AEON-7, MLX, MLX-LM, and MTPLX.