Instructions to use AutomatosX/AX-Tiel-Coder-35B-A3B-MLX-AXQ-6bit-MTP with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use AutomatosX/AX-Tiel-Coder-35B-A3B-MLX-AXQ-6bit-MTP with MLX:

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

# Generate text with mlx-lm
from mlx_lm import load, generate

model, tokenizer = load("AutomatosX/AX-Tiel-Coder-35B-A3B-MLX-AXQ-6bit-MTP")

prompt = "Write a story about Einstein"
messages = [{"role": "user", "content": prompt}]
prompt = tokenizer.apply_chat_template(
    messages, add_generation_prompt=True
)

text = generate(model, tokenizer, prompt=prompt, verbose=True)

Notebooks
Google Colab
Kaggle
Local Apps Settings
LM Studio

How to use AutomatosX/AX-Tiel-Coder-35B-A3B-MLX-AXQ-6bit-MTP with Pi:

Start the MLX server

# Install MLX LM:
uv tool install mlx-lm
# Start a local OpenAI-compatible server:
mlx_lm.server --model "AutomatosX/AX-Tiel-Coder-35B-A3B-MLX-AXQ-6bit-MTP"

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": "AutomatosX/AX-Tiel-Coder-35B-A3B-MLX-AXQ-6bit-MTP"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

MLX LM

How to use AutomatosX/AX-Tiel-Coder-35B-A3B-MLX-AXQ-6bit-MTP with MLX LM:

Generate or start a chat session

# Install MLX LM
uv tool install mlx-lm
# Interactive chat REPL
mlx_lm.chat --model "AutomatosX/AX-Tiel-Coder-35B-A3B-MLX-AXQ-6bit-MTP"

Run an OpenAI-compatible server

# Install MLX LM
uv tool install mlx-lm
# Start the server
mlx_lm.server --model "AutomatosX/AX-Tiel-Coder-35B-A3B-MLX-AXQ-6bit-MTP"
# Calling the OpenAI-compatible server with curl
curl -X POST "http://localhost:8000/v1/chat/completions" \
   -H "Content-Type: application/json" \
   --data '{
     "model": "AutomatosX/AX-Tiel-Coder-35B-A3B-MLX-AXQ-6bit-MTP",
     "messages": [
       {"role": "user", "content": "Hello"}
     ]
   }'

Hermes Agent

How to use AutomatosX/AX-Tiel-Coder-35B-A3B-MLX-AXQ-6bit-MTP 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 "AutomatosX/AX-Tiel-Coder-35B-A3B-MLX-AXQ-6bit-MTP"

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 AutomatosX/AX-Tiel-Coder-35B-A3B-MLX-AXQ-6bit-MTP

Run Hermes

hermes

Atomic Chat

OpenClaw

How to use AutomatosX/AX-Tiel-Coder-35B-A3B-MLX-AXQ-6bit-MTP with OpenClaw:

Start the MLX server

# Install MLX LM:
uv tool install mlx-lm
# Start a local OpenAI-compatible server:
mlx_lm.server --model "AutomatosX/AX-Tiel-Coder-35B-A3B-MLX-AXQ-6bit-MTP"

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 "AutomatosX/AX-Tiel-Coder-35B-A3B-MLX-AXQ-6bit-MTP" \
  --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"

AX-Tiel-Coder-35B-A3B-MLX-AXQ-6bit-MTP

Research development artifact, not certified. Affine six-bit base packing with higher-precision protection; this is not MXFP6 or a uniformly six-bit checkpoint.

Re-quantized from an already quantized oQ6e checkpoint. The source Sharp chat template and protected MTP payload are retained. This conversion does not recover BF16 information already lost in the source, reproduce its calibration process, or inherit upstream GGUF, SWE-bench, quality or MTP-speed results.

Property	Value
Source	peculiar-ragdoll/Tiel-Coder-35B-A3B-MLX-oQ6e-MTP
Immutable source revision	`88625754ac91b542280a5602239ce6b2166366f0`
Manifest budget class	`7p5bpw`
Complete measured storage	7.490084 BPW
Weight storage	33,660,273,640 bytes / 31.3486 GiB
MTP	Protected payload included; execution, acceptance and acceleration unverified
Quality	No quality benchmark or certificate for this revision

Precision distribution

The following shares are logical parameters, not file-byte shares.

Precision	Parameter share	Parameters
6bit	93.0325%	33,446,887,424
8bit	1.9582%	704,004,096
bf16	5.0093%	1,800,931,184

Group sizes and per-module overrides are recorded in config.json and axquant_plan.json. Weight bytes do not establish peak runtime memory requirements. All advertised BPW values include protection and storage overhead.

Runtime evidence and limits

MLX-VLM 0.7.9 / MLX 0.32.4 passed two short text smokes on Device(gpu, 0): arithmetic produced 42 and exact-word output produced READY. The text smoke used the public MLX-VLM load and generate APIs on Metal.

These checks establish only scoped text load/generation. They do not establish coding quality, long-context behavior, vision/video quality, MTP execution or speed. Vision and MTP weights are present; neither feature is behavior-verified by this release. The -MTP suffix describes packaged weights, not a measured acceleration claim. The tested text loading path disables the unindexed vision tower. Image inference is not supported by that tested path for this revision; the retained vision sidecar is not a verified vision runtime export.

Runtime	Status for this artifact
MLX-VLM	Scoped text smoke as described above
MLX-LM	Not tested; do not infer equivalence from successful loading
AX Engine	Not measured; an included native manifest is not execution evidence
omlx	unsupported; no load, generation or MTP claim for this artifact
mtplx	unsupported; no load, generation or MTP claim for this artifact

See the manifest-bound runtime compatibility record. Any runtime-specific index, norm or n-gram layout export is a distinct artifact and needs its own load/generation validation. No peer variant is silently substituted here.

Download and evaluation

python -m pip install -U huggingface_hub mlx-vlm
hf download AutomatosX/AX-Tiel-Coder-35B-A3B-MLX-AXQ-6bit-MTP --local-dir ./AX-Tiel-Coder-35B-A3B-MLX-AXQ-6bit-MTP

Pin the resulting immutable Hub commit for reproducibility. Use a runtime build that supports the exact source architecture and declared physical quantization. Do not assume a particular memory capacity, modality or MTP setting is supported.

Provenance

The manifest binds model files, precision plan, quantizer execution and runtime compatibility checks. Protected MTP provenance is in axquant_mtp_sidecar_manifest.json. Source certificates, upstream benchmark plates, and measurements from other AXQ models do not transfer to this checkpoint.

License and attribution

The pinned upstream repository declares MIT in its model-card metadata. Retain the upstream attribution and consult its model card for provenance and terms. Upstream attribution: peculiar-ragdoll and ornith-ai.

Downloads last month: 42

Safetensors

Model size

35B params

Tensor type

BF16

U32

MLX

Hardware compatibility

6-bit

Model tree for AutomatosX/AX-Tiel-Coder-35B-A3B-MLX-AXQ-6bit-MTP

Base model

ornith-ai/Ornith-1.5-35B-A3B

Quantized

peculiar-ragdoll/Tiel-Coder-35B-A3B-MLX-oQ6e-MTP

Quantized

(3)

this model

Collections including AutomatosX/AX-Tiel-Coder-35B-A3B-MLX-AXQ-6bit-MTP