Image-Text-to-Text
MLX
Safetensors
modilify_mk1
diffusion
multimodal
mixture-of-experts
conversational
Instructions to use modilify/Modilify-Mk1-MLX with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- MLX
How to use modilify/Modilify-Mk1-MLX 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("modilify/Modilify-Mk1-MLX") config = load_config("modilify/Modilify-Mk1-MLX") # 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
- Pi
How to use modilify/Modilify-Mk1-MLX with Pi:
Start the MLX server
# Install MLX LM: uv tool install mlx-lm # Start a local OpenAI-compatible server: mlx_lm.server --model "modilify/Modilify-Mk1-MLX"
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": "modilify/Modilify-Mk1-MLX" } ] } } }Run Pi
# Start Pi in your project directory: pi
- Hermes Agent
How to use modilify/Modilify-Mk1-MLX 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 "modilify/Modilify-Mk1-MLX"
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 modilify/Modilify-Mk1-MLX
Run Hermes
hermes
- Atomic Chat
- OpenClaw
How to use modilify/Modilify-Mk1-MLX with OpenClaw:
Start the MLX server
# Install MLX LM: uv tool install mlx-lm # Start a local OpenAI-compatible server: mlx_lm.server --model "modilify/Modilify-Mk1-MLX"
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 "modilify/Modilify-Mk1-MLX" \ --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"
File size: 6,118 Bytes
a066584 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 | #!/usr/bin/env python3
# Copyright 2026 Modilify
# SPDX-License-Identifier: LicenseRef-Modilify-Open-Model-1.0
"""Convert an official Modilify Mk1 checkpoint into native MLX shards."""
from __future__ import annotations
import argparse
import json
import shutil
import sys
from pathlib import Path
import mlx.core as mx
ROOT = Path(__file__).resolve().parent
if str(ROOT) not in sys.path:
sys.path.insert(0, str(ROOT))
from modilify_mlx.config import MODEL_TYPE, TEXT_MODEL_TYPE, ModilifyMk1Config
from modilify_mlx.convert_utils import remap_weight, should_keep_source_key
COPY_PATTERNS = (
"tokenizer.json",
"tokenizer_config.json",
"chat_template.jinja",
"processor_config.json",
"generation_config.json",
"special_tokens_map.json",
"preprocessor_config.json",
)
MAX_SHARD_BYTES = 5 * 1024**3
def _load_json(path: Path) -> dict:
with path.open(encoding="utf-8") as handle:
return json.load(handle)
def _write_config(source: Path, destination: Path) -> None:
payload = _load_json(source / "config.json")
if payload.get("model_type") != MODEL_TYPE:
raise ValueError(
f"Source model_type must be {MODEL_TYPE!r}, got {payload.get('model_type')!r}."
)
payload.pop("auto_map", None)
payload["architectures"] = ["ModilifyMk1ForBlockDiffusion"]
text = dict(payload.get("text_config") or {})
text["model_type"] = TEXT_MODEL_TYPE
payload["text_config"] = text
payload["model_type"] = MODEL_TYPE
# Validate through the native config so a bad export fails here.
ModilifyMk1Config.from_dict(payload)
with (destination / "config.json").open("w", encoding="utf-8") as handle:
json.dump(payload, handle, indent=2)
handle.write("\n")
def _copy_sidecar_files(source: Path, destination: Path) -> None:
for name in COPY_PATTERNS:
src = source / name
if src.exists():
shutil.copy2(src, destination / name)
def _flush_shard(
destination: Path,
shard: dict[str, mx.array],
shard_index: int,
planned_count: int,
weight_map: dict[str, str],
) -> int:
if not shard:
return shard_index
name = f"model-{shard_index:05d}-of-{planned_count:05d}.safetensors"
mx.save_safetensors(
str(destination / name),
shard,
metadata={"format": "mlx"},
)
for key in shard:
weight_map[key] = name
return shard_index + 1
def convert(source: Path, destination: Path) -> None:
source = source.resolve()
destination = destination.resolve()
destination.mkdir(parents=True, exist_ok=True)
index = _load_json(source / "model.safetensors.index.json")
source_map: dict[str, str] = index["weight_map"]
shard_names = []
for name in source_map.values():
if name not in shard_names:
shard_names.append(name)
remapped: dict[str, mx.array] = {}
kept_source = 0
dropped_source = 0
for shard_name in shard_names:
print(f"[convert] reading {shard_name}", flush=True)
loaded = mx.load(str(source / shard_name))
for key, value in loaded.items():
if not should_keep_source_key(key):
dropped_source += 1
continue
kept_source += 1
for new_key, new_value in remap_weight(key, value):
if new_key in remapped:
raise ValueError(f"Duplicate remapped key: {new_key}")
remapped[new_key] = new_value
del loaded
print(
f"[convert] kept {kept_source} source tensors, dropped {dropped_source}, "
f"wrote {len(remapped)} MLX tensors",
flush=True,
)
latent_keys = [key for key in remapped if key.startswith("latent_deliberation.")]
if len(latent_keys) < 150:
raise RuntimeError(
f"Expected the latent stack to survive conversion, found {len(latent_keys)} keys."
)
items = sorted(remapped.items())
shards: list[dict[str, mx.array]] = []
current: dict[str, mx.array] = {}
current_bytes = 0
for key, value in items:
tensor_bytes = int(value.nbytes)
if current and current_bytes + tensor_bytes > MAX_SHARD_BYTES:
shards.append(current)
current = {}
current_bytes = 0
current[key] = value
current_bytes += tensor_bytes
if current:
shards.append(current)
planned = max(1, len(shards))
weight_map: dict[str, str] = {}
total_size = 0
for index_i, shard in enumerate(shards, start=1):
print(f"[convert] writing shard {index_i}/{planned}", flush=True)
name = (
f"model-{index_i:05d}-of-{planned:05d}.safetensors"
if planned > 1
else "model.safetensors"
)
mx.save_safetensors(
str(destination / name),
shard,
metadata={"format": "mlx"},
)
for key, value in shard.items():
weight_map[key] = name
total_size += int(value.nbytes)
index_payload = {
"metadata": {"total_size": total_size},
"weight_map": {key: weight_map[key] for key in sorted(weight_map)},
}
with (destination / "model.safetensors.index.json").open(
"w", encoding="utf-8"
) as handle:
json.dump(index_payload, handle, indent=2)
handle.write("\n")
_write_config(source, destination)
_copy_sidecar_files(source, destination)
print(f"[convert] done -> {destination}", flush=True)
print(f"[convert] tensors={len(weight_map)} bytes={total_size}", flush=True)
def main() -> None:
parser = argparse.ArgumentParser(description=__doc__)
parser.add_argument(
"--source",
type=Path,
default=Path.home() / "Modilify-Mk1",
help="Official Mk1 Hugging Face directory",
)
parser.add_argument(
"--destination",
type=Path,
default=ROOT,
help="Native MLX output directory",
)
args = parser.parse_args()
convert(args.source, args.destination)
if __name__ == "__main__":
main()
|