Image-Text-to-Text
MLX
Safetensors
deepseek_v41
gbx-lm
apple-silicon
mixture-of-experts
4-bit precision
expert-paging
fp8
Instructions to use GreenBitAI/DeepSeek-V4.1-Flash-4bit-paged with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- MLX
How to use GreenBitAI/DeepSeek-V4.1-Flash-4bit-paged 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("GreenBitAI/DeepSeek-V4.1-Flash-4bit-paged") config = load_config("GreenBitAI/DeepSeek-V4.1-Flash-4bit-paged") # 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
- Atomic Chat
Commit History
9 days ago
-
Describe image input
Verified signature098043c
12 days ago
-
Add DGX Spark support to the card
Verified signatured7ad585
14 days ago
-
Update draft-head speed tables (v0.8.0)
Verified signature9c88c27
26 days ago
-
State library_name: mlx
Verified signature41944e9 -
Document gbx_lm generate and chat (v0.7.1)
Verified signature3c04a36
28 days ago
-
Tag the card gbx-lm
Verified signature92eb57b
30 days ago
-
Model card: drop the duplicated command lines
Verified signatureTwo different python -m invocations became the same gbx_lm line, and were left printed twice.5e574f9 -
Model card: clear the old unpack directory when upgrading
Verified signatureA second version unpacks into the same directory as the first, and macOS kills a process that mixes two builds -- `Killed: 9`, with nothing to say why. The install now clears it, and both that and the PATH case are written out.6c97420
-
Model card: say what a smaller machine trades for fitting
Verified signaturePaging already said a machine that cannot hold the experts streams them. This says what that costs -- a wait on the SSD per token that routes to an expert not in memory -- without inventing a number we have not measured.9723a84 -
Model card: use the server's own default port
Verified signature11688 is what gbx_lm listens on without being told, so the run command no longer passes --port and every example uses that port.9961411 -
Model card: point Codex and Claude Code at the same server
Verified signatureThe server answers /v1/chat/completions, /v1/responses and /v1/messages on one port; the card now shows the provider and settings each client needs.5e52de1 -
Model card: install and run the signed gbx_lm binary
Verified signatureThe Python package is no longer published; the card now documents the notarised Apple Silicon binary, its macOS and chip requirements, and the memory this build actually takes.13c1842
-
Draft-head numbers from the 2026-09-18 run
Verified signature49fd809 -
Describe the draft head
Verified signature3f8f270
-
Add files using upload-large-folder tool
Verified signature2e2fc8a -
Add files using upload-large-folder tool
Verified signature706ca3c -
initial commit
Verified signaturedb8d251