--- license: apache-2.0 license_link: https://www.apache.org/licenses/LICENSE-2.0 library_name: mlx pipeline_tag: image-text-to-text base_model: thinkingmachines/Inkling-2-Small tags: - moe - mlx --- # mlx-community/Inkling-2-Small-4bit This model [mlx-community/Inkling-2-Small-4bit](https://huggingface.co/mlx-community/Inkling-2-Small-4bit) was converted to MLX format from [thinkingmachines/Inkling-2-Small](https://huggingface.co/thinkingmachines/Inkling-2-Small) using mlx-lm with [ml-explore/mlx-lm#2061](https://github.com/ml-explore/mlx-lm/pull/2061). Refer to the [original model card](https://huggingface.co/thinkingmachines/Inkling-2-Small) for more details on the model. It takes text and image input. The vision tower is included. The DSpark draft is not included. The routed experts keep the original quantization. They are NVFP4 with a per-expert global scale in layers 0 to 39 and MXFP8 in layers 40 to 47. The attention, embedding and output head weights are 4-bit affine with group size 64. The vision tower stays in bfloat16 and the router in float32. Tested with the branch of that pull request. ## Use with mlx ```bash pip install -U git+https://github.com/ml-explore/mlx-lm.git@refs/pull/2061/head ``` ```python from mlx_lm import load, generate model, tokenizer = load("mlx-community/Inkling-2-Small-4bit") prompt = "hello" if tokenizer.chat_template is not None: messages = [{"role": "user", "content": prompt}] prompt = tokenizer.apply_chat_template( messages, add_generation_prompt=True, return_dict=False, ) response = generate(model, tokenizer, prompt=prompt, verbose=True) ``` For image input, run `mlx_lm.server` and send the images as `image_url` parts of a chat message.