Instructions to use MedMLX/MLX-Reason-CT with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use MedMLX/MLX-Reason-CT 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("MedMLX/MLX-Reason-CT")
config = load_config("MedMLX/MLX-Reason-CT")

# 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 MedMLX/MLX-Reason-CT with Pi:

Start the MLX server

# Install MLX LM:
uv tool install mlx-lm
# Start a local OpenAI-compatible server:
mlx_lm.server --model "MedMLX/MLX-Reason-CT"

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": "MedMLX/MLX-Reason-CT"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Hermes Agent

How to use MedMLX/MLX-Reason-CT 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 "MedMLX/MLX-Reason-CT"

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 MedMLX/MLX-Reason-CT

Run Hermes

hermes

Atomic Chat

OpenClaw

How to use MedMLX/MLX-Reason-CT with OpenClaw:

Start the MLX server

# Install MLX LM:
uv tool install mlx-lm
# Start a local OpenAI-compatible server:
mlx_lm.server --model "MedMLX/MLX-Reason-CT"

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 "MedMLX/MLX-Reason-CT" \
  --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"

MLX-Reason-CT / reference-comparison.json

josand

Upload converted NV-Reason-CT weights and release metadata

56baf43 verified 22 days ago

Raw History Blame Contribute Delete

2.65 kB

{
  "schema_version": "medmlx.reference.v1",
  "model": "nv-reason-ct",
  "date": "2026-10-12",
  "case": {
    "dataset": "LUNA16 subset0",
    "case_id": "1.3.6.1.4.1.14519.5.2.1.6279.6001.105756658031515062000744821260",
    "input_sha256": "5a7c3a4e681dcba5585bb49e2f355cbc08311b2944614003966409f9b6a71dfe"
  },
  "original": {
    "code": "nvidia/NV-Reason-CT (Hugging Face remote model.py)",
    "version_or_commit": "386b93e034983f6c1fc841a43833a1b6a0cd9c13",
    "device": "cpu",
    "precision": "FP32",
    "checkpoint_sha256": "655c15e59dfa1957f7d5f475b1702d689fb7107160f21256028391bb5f6e6ba5",
    "command": "PYTHONDONTWRITEBYTECODE=1 work/reference/nv-reason-ct/.venv/bin/python benchmarks/reference/nv-reason-ct.py",
    "mode": "greedy_generation",
    "duration_s": 1244.5534047500114,
    "environment": {
      "torch": "2.13.0",
      "transformers": "5.6.2",
      "dynamic-network-architectures": "0.4.3",
      "timm": "1.0.22",
      "monai": "1.6.0rc1",
      "numpy": "2.3.5",
      "nibabel": "5.4.2"
    }
  },
  "mlx": {
    "run_record": "results/runs/nv-reason-ct.json",
    "output_path": "<paper-repo>/work/runs/nv-reason-ct/out/run.json",
    "output_sha256": "a9af3ecaf7d50ee2669a10699a773bc96274cef4aa073885eb6fcf2ad6b20fb9",
    "report": {
      "path": "<paper-repo>/work/runs/nv-reason-ct/out/report.txt",
      "sha256": "06d5943d8465cbda9743273566eb8ac802ebf44e0621f08a83f7eb78977bc282"
    },
    "preprocessing_revision": "eb2c10baf688b32eee082dbb8973e88b4555db7a"
  },
  "comparison": {
    "metric": "token_identical",
    "value": true,
    "threshold": true,
    "result": "pass",
    "first_divergence": null,
    "original_generated_tokens": 208,
    "mlx_generated_tokens": 208,
    "original_terminated_by_eos": true,
    "original_truncated_by_max_new_tokens": false
  },
  "preprocessing": {
    "max_abs_diff": 9.375810623168945e-05,
    "prompt_tokens_identical": true,
    "pixel_input": "recorded MLX revision's deterministic host preprocessing"
  },
  "notes": "Same SHA-256 verified LUNA16 NIfTI and original model.safetensors. Source snapshot relocated from radnn to research/radnn; source/model/template/tokenizer files match the recorded pins. Same prompt: write a structured chest CT report. Thinking off, greedy, max_new_tokens 512, FP32 CPU, SDPA, cache enabled. Original processor/template and prompt tokens are checked against recorded native host preprocessing; the identical recorded host-preprocessed voxels are explicitly supplied to the original network. Saved MLX output tokens come directly from its hash-verified run.json. No MLX generation rerun.",
  "failure": null
}

	{
	"schema_version": "medmlx.reference.v1",
	"model": "nv-reason-ct",
	"date": "2026-10-12",
	"case": {
	"dataset": "LUNA16 subset0",
	"case_id": "1.3.6.1.4.1.14519.5.2.1.6279.6001.105756658031515062000744821260",
	"input_sha256": "5a7c3a4e681dcba5585bb49e2f355cbc08311b2944614003966409f9b6a71dfe"
	},
	"original": {
	"code": "nvidia/NV-Reason-CT (Hugging Face remote model.py)",
	"version_or_commit": "386b93e034983f6c1fc841a43833a1b6a0cd9c13",
	"device": "cpu",
	"precision": "FP32",
	"checkpoint_sha256": "655c15e59dfa1957f7d5f475b1702d689fb7107160f21256028391bb5f6e6ba5",
	"command": "PYTHONDONTWRITEBYTECODE=1 work/reference/nv-reason-ct/.venv/bin/python benchmarks/reference/nv-reason-ct.py",
	"mode": "greedy_generation",
	"duration_s": 1244.5534047500114,
	"environment": {
	"torch": "2.13.0",
	"transformers": "5.6.2",
	"dynamic-network-architectures": "0.4.3",
	"timm": "1.0.22",
	"monai": "1.6.0rc1",
	"numpy": "2.3.5",
	"nibabel": "5.4.2"
	}
	},
	"mlx": {
	"run_record": "results/runs/nv-reason-ct.json",
	"output_path": "<paper-repo>/work/runs/nv-reason-ct/out/run.json",
	"output_sha256": "a9af3ecaf7d50ee2669a10699a773bc96274cef4aa073885eb6fcf2ad6b20fb9",
	"report": {
	"path": "<paper-repo>/work/runs/nv-reason-ct/out/report.txt",
	"sha256": "06d5943d8465cbda9743273566eb8ac802ebf44e0621f08a83f7eb78977bc282"
	},
	"preprocessing_revision": "eb2c10baf688b32eee082dbb8973e88b4555db7a"
	},
	"comparison": {
	"metric": "token_identical",
	"value": true,
	"threshold": true,
	"result": "pass",
	"first_divergence": null,
	"original_generated_tokens": 208,
	"mlx_generated_tokens": 208,
	"original_terminated_by_eos": true,
	"original_truncated_by_max_new_tokens": false
	},
	"preprocessing": {
	"max_abs_diff": 9.375810623168945e-05,
	"prompt_tokens_identical": true,
	"pixel_input": "recorded MLX revision's deterministic host preprocessing"
	},
	"notes": "Same SHA-256 verified LUNA16 NIfTI and original model.safetensors. Source snapshot relocated from radnn to research/radnn; source/model/template/tokenizer files match the recorded pins. Same prompt: write a structured chest CT report. Thinking off, greedy, max_new_tokens 512, FP32 CPU, SDPA, cache enabled. Original processor/template and prompt tokens are checked against recorded native host preprocessing; the identical recorded host-preprocessed voxels are explicitly supplied to the original network. Saved MLX output tokens come directly from its hash-verified run.json. No MLX generation rerun.",
	"failure": null
	}