Instructions to use Hygaard/Nex-N2.6-oQ4e-mtp with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- MLX
How to use Hygaard/Nex-N2.6-oQ4e-mtp 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("Hygaard/Nex-N2.6-oQ4e-mtp") config = load_config("Hygaard/Nex-N2.6-oQ4e-mtp") # 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 Hygaard/Nex-N2.6-oQ4e-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 "Hygaard/Nex-N2.6-oQ4e-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": "Hygaard/Nex-N2.6-oQ4e-mtp" } ] } } }Run Pi
# Start Pi in your project directory: pi
- Hermes Agent
How to use Hygaard/Nex-N2.6-oQ4e-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 "Hygaard/Nex-N2.6-oQ4e-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 Hygaard/Nex-N2.6-oQ4e-mtp
Run Hermes
hermes
- Atomic Chat
- OpenClaw
How to use Hygaard/Nex-N2.6-oQ4e-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 "Hygaard/Nex-N2.6-oQ4e-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 "Hygaard/Nex-N2.6-oQ4e-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"
Nex-N2.6-oQ4e-mtp
oQ4e mixed-precision MLX conversion of Nex-AGI's Nex-N2.6, intended for Apple Silicon Macs with at least 256 GB of unified memory.
Effective weight storage: 4.64 BPW ยท 186.40 GB ยท group size 64.
Nex-N2.6 is a multimodal agentic model based on GLM-5.3-Flash, with approximately 321B parameters. This community conversion retains the full expert set, vision encoder and multi-token-prediction head. No additional training, abliteration or expert pruning was applied.
Conversion
- Source: creator FP8 checkpoint,
nex-agi/Nex-N2.6@987472298e7867721576ac49a253a6f5b526013e. - Format: MLX safetensors; oQ4e mixed precision, affine packing, group size 64.
- Tooling: oMLX
0.7.4.dev1native oQe streaming converter. - Calibration: 256 samples, 512 tokens each, using the bundled code/multilingual corpus; expert-coverage and structural checks passed.
- Weights: 35 shards, 186.40 GB / 173.60 GiB. Runtime and context state need additional memory.
Detailed conversion and preservation records are in CONVERSION.json.
Mixed precision
4.64 effective bits per weight (BPW). The packed quantized matrices alone average 4.112 bits per weight; the effective figure includes their BF16 scales/biases, every unquantized tensor and safetensors headers.
| Component | Weight precision | Original parameters | Stored GB |
|---|---|---|---|
| Routed MoE experts | 4-bit | 311,653,564,416 | 175.305130 |
| Attention projections | 8-bit | 6,171,918,336 | 6.557663 |
| Shared experts | 8-bit | 1,082,130,432 | 1.149764 |
| Dense MLP projections | 8-bit | 452,984,832 | 0.481296 |
| Token embeddings | 8-bit | 634,388,480 | 0.674038 |
| Output head | 8-bit | 634,388,480 | 0.674038 |
| Vision encoder | BF16 | 563,627,008 | 1.127254 |
| Linear-attention convolutions | BF16 | 3,342,336 | 0.006685 |
| Indexer pooling/compression | BF16 | 6,297,600 | 0.012595 |
| Text/MTP normalization | BF16 | 425,216 | 0.000850 |
| MTP fusion projection | BF16 | 33,554,432 | 0.067109 |
| MoE routers/correction biases | FP32 | 50,737,248 | 0.202949 |
| Hyperconnections | FP32 | 35,391,870 | 0.141567 |
| Recurrent decay controls | FP32 | 280,704 | 0.001123 |
| Safetensors headers | โ | โ | 0.000403 |
| Total | 4.64 effective BPW | 321,323,031,390 | 186.402464 |
Stored GB uses decimal units. Quantized rows include their BF16 affine scales/biases; normalization in this table excludes the vision encoder, whose tensors are all counted in the vision row. Routed/shared expert and attention totals include MTP; the MTP fusion projection is listed separately. No tensors are counted twice.
Calculation: 8 ร 186,402,464,045 stored bytes รท 321,323,031,390 original scalar parameters = 4.6409 BPW. Affine metadata contributes storage but is not counted as additional original model parameters. Tokenizer/config files, KV cache and runtime workspace are excluded.
Per-block precision map
| Block | Routed expert matrices | Other quantized matrices | Unquantized tensors | Stored GB |
|---|---|---|---|---|
| Text 0 | โ (dense MLP) | 8-bit | BF16, FP32 | 0.310052 |
| Text 1 | โ (dense MLP) | 8-bit | BF16, FP32 | 0.310052 |
| Text 2 | โ (dense MLP) | 8-bit | BF16, FP32 | 0.310052 |
| Text 3 | 4-bit | 8-bit | BF16, FP32 | 4.244700 |
| Text 4 | 4-bit | 8-bit | BF16, FP32 | 4.257941 |
| Text 5 | 4-bit | 8-bit | BF16, FP32 | 4.257941 |
| Text 6 | 4-bit | 8-bit | BF16, FP32 | 4.257941 |
| Text 7 | 4-bit | 8-bit | BF16, FP32 | 4.244700 |
| Text 8 | 4-bit | 8-bit | BF16, FP32 | 4.257941 |
| Text 9 | 4-bit | 8-bit | BF16, FP32 | 4.257941 |
| Text 10 | 4-bit | 8-bit | BF16, FP32 | 4.257941 |
| Text 11 | 4-bit | 8-bit | BF16, FP32 | 4.244700 |
| Text 12 | 4-bit | 8-bit | BF16, FP32 | 4.257941 |
| Text 13 | 4-bit | 8-bit | BF16, FP32 | 4.257941 |
| Text 14 | 4-bit | 8-bit | BF16, FP32 | 4.257941 |
| Text 15 | 4-bit | 8-bit | BF16, FP32 | 4.244700 |
| Text 16 | 4-bit | 8-bit | BF16, FP32 | 4.257941 |
| Text 17 | 4-bit | 8-bit | BF16, FP32 | 4.257941 |
| Text 18 | 4-bit | 8-bit | BF16, FP32 | 4.257941 |
| Text 19 | 4-bit | 8-bit | BF16, FP32 | 4.244700 |
| Text 20 | 4-bit | 8-bit | BF16, FP32 | 4.257941 |
| Text 21 | 4-bit | 8-bit | BF16, FP32 | 4.257941 |
| Text 22 | 4-bit | 8-bit | BF16, FP32 | 4.257941 |
| Text 23 | 4-bit | 8-bit | BF16, FP32 | 4.244700 |
| Text 24 | 4-bit | 8-bit | BF16, FP32 | 4.257941 |
| Text 25 | 4-bit | 8-bit | BF16, FP32 | 4.257941 |
| Text 26 | 4-bit | 8-bit | BF16, FP32 | 4.257941 |
| Text 27 | 4-bit | 8-bit | BF16, FP32 | 4.244700 |
| Text 28 | 4-bit | 8-bit | BF16, FP32 | 4.257941 |
| Text 29 | 4-bit | 8-bit | BF16, FP32 | 4.257941 |
| Text 30 | 4-bit | 8-bit | BF16, FP32 | 4.257941 |
| Text 31 | 4-bit | 8-bit | BF16, FP32 | 4.244700 |
| Text 32 | 4-bit | 8-bit | BF16, FP32 | 4.257941 |
| Text 33 | 4-bit | 8-bit | BF16, FP32 | 4.257941 |
| Text 34 | 4-bit | 8-bit | BF16, FP32 | 4.257941 |
| Text 35 | 4-bit | 8-bit | BF16, FP32 | 4.244700 |
| Text 36 | 4-bit | 8-bit | BF16, FP32 | 4.257941 |
| Text 37 | 4-bit | 8-bit | BF16, FP32 | 4.257941 |
| Text 38 | 4-bit | 8-bit | BF16, FP32 | 4.257941 |
| Text 39 | 4-bit | 8-bit | BF16, FP32 | 4.244700 |
| Text 40 | 4-bit | 8-bit | BF16, FP32 | 4.257941 |
| Text 41 | 4-bit | 8-bit | BF16, FP32 | 4.257941 |
| Text 42 | 4-bit | 8-bit | BF16, FP32 | 4.257941 |
| Text 43 | 4-bit | 8-bit | BF16, FP32 | 4.244700 |
| Text 44 | 4-bit | 8-bit | BF16, FP32 | 4.257941 |
| MTP 0 | 4-bit | 8-bit | BF16, FP32 | 4.308687 |
Complete per-module/tensor precision data (CSV) and exact accounting/methodology (JSON) are included. The CSV covers every quantized module and unquantized tensor; quantized module rows include their scales and affine biases.
Running
Use a runtime supporting Nex-N2.6's glm5_next architecture and oQe mixed precision, with the included tokenizer, processor and chat template. The template supports reasoning_effort values low, high and max (default).
Cold loading and text generation tested on a 256 GB Apple Silicon Mac with oMLX 0.7.4.dev1; MTP decoding active. Broader quality, vision and long-context testing is ongoing.
Other precision variants
- Nex-N2.6 oQ3e: 3.69 effective BPW, 148.35 GB; intended for 192 GB Apple Silicon Macs (M2 execution untested).
- Nex-N2.6 oQ5e: 5.61 effective BPW, 225.36 GB. Both variants are converted directly from the same pinned creator FP8 checkpoint, not requantized from one another.
Credits and license
Model: Nex-AGI. GLM base: Z.ai. Conversion tooling: oMLX / oQe. Community conversion: Hygaard.
Licensed under Apache-2.0, with the upstream GLM MIT notice retained. See NOTICE for attribution and modification notes.
- Downloads last month
- 133
4-bit
Model tree for Hygaard/Nex-N2.6-oQ4e-mtp
Base model
nex-agi/Nex-N2.6