📋CLI Reference

Nexa commands overview

Overview

usage: nexa [-h] [-V] {run,onnx,server,pull,remove,clean,list,login,whoami,logout} ...

Nexa CLI tool for handling various model operations.

positional arguments:
  {run,onnx,server,pull,remove,clean,list,login,whoami,logout}
                        sub-command help
    run                 Run inference for various tasks using GGUF models.
    onnx                Run inference for various tasks using ONNX models.
    embed               Generate embeddings for text.
    server              Run the Nexa AI Text Generation Service
    pull                Pull a model from official or hub.
    remove              Remove a model from local machine.
    clean               Clean up all model files.
    list                List all models in the local machine.
    login               Login to Nexa API.
    whoami              Show current user information.
    logout              Logout from Nexa API.

options:
  -h, --help            show this help message and exit
  -V, --version         Show the version of the Nexa SDK.

List local models

List all models on your local computer.

nexa list

Download a model

Download a model file to your local computer from Nexa Model Hub.

nexa pull MODEL_PATH
usage: nexa pull [-h] model_path

positional arguments:
  model_path  Path or identifier for the model in Nexa Model Hub

options:
  -h, --help  show this help message and exit
  -hf, --huggingface  Pull model from Hugging Face Hub

Example Command:

nexa pull llama2

Remove a model

Remove a model from your local computer.

nexa remove MODEL_PATH
usage: nexa remove [-h] model_path

positional arguments:
  model_path  Path or identifier for the model in Nexa Model Hub

options:
  -h, --help  show this help message and exit

Example Command:

nexa remove llama2

Remove all downloaded models

Remove all downloaded models on your local computer.

nexa clean

Run a model

Run a model on your local computer. If the model file is not yet downloaded, it will be automatically fetched first. For more details, please refer to the Inference page.

By default nexa run will run GGUF models. Use nexa onnx to run ONNX models.

Remember to install nexaai onnx version in advance.

Run text-generation model

Run text-generation models on your local computer.

nexa run MODEL_PATH
usage: nexa run [-h] [-t TEMPERATURE] [-m MAX_NEW_TOKENS] [-k TOP_K] [-p TOP_P] [-sw [STOP_WORDS ...]] [-pf] [-st] model_path

positional arguments:
  model_path            Path or identifier for the model in Nexa Model Hub

options:
  -h, --help            show this help message and exit
  -pf, --profiling      Enable profiling logs for the inference process
  -st, --streamlit      Run the inference in Streamlit UI
  -lp, --local_path     Indicate that the model path provided is the local path, must be used with -mt
  -mt, --model_type     Indicate the model running type, must be used with -lp or -hf, choose from [NLP, COMPUTER_VISION, MULTIMODAL, AUDIO]
  -hf, --huggingface    Load model from Hugging Face Hub, must be used with -mt

Text generation options:
  -t, --temperature TEMPERATURE
                        Temperature for sampling
  -m, --max_new_tokens MAX_NEW_TOKENS
                        Maximum number of new tokens to generate
  -k, --top_k TOP_K     Top-k sampling parameter
  -p, --top_p TOP_P     Top-p sampling parameter
  -sw, --stop_words [STOP_WORDS ...]
                        List of stop words for early stopping
  --lora_path           Path to a LoRA file to apply to the model                        
  --nctx                Maximum context length of the model you're using

Example Command:

nexa run llama2

Run image-generation model

Run image-generation models on your local computer.

nexa run MODEL_PATH
usage: nexa run [-h] [-i2i] [-ns NUM_INFERENCE_STEPS] [-np NUM_IMAGES_PER_PROMPT] [-H HEIGHT] [-W WIDTH] [-g GUIDANCE_SCALE] [-o OUTPUT] [-s RANDOM_SEED] [-st] model_path

positional arguments:
  model_path            Path or identifier for the model in Nexa Model Hub

options:
  -h, --help            show this help message and exit
  -st, --streamlit      Run the inference in Streamlit UI, can be used with -lp or -hf
  -lp, --local_path     Indicate that the model path provided is the local path, must be used with -mt
  -mt, --model_type     Indicate the model running type, must be used with -lp or -hf, choose from [NLP, COMPUTER_VISION, MULTIMODAL, AUDIO]
  -hf, --huggingface    Load model from Hugging Face Hub, must be used with -mt
  
Image generation options:
  -i2i, --img2img       Whether to run image-to-image generation
  -ns, --num_inference_steps NUM_INFERENCE_STEPS
                        Number of inference steps
  -np, --num_images_per_prompt NUM_IMAGES_PER_PROMPT
                        Number of images to generate per prompt
  -H, --height HEIGHT   Height of the output image
  -W, --width WIDTH     Width of the output image
  -g, --guidance_scale GUIDANCE_SCALE
                        Guidance scale for diffusion
  -o, --output OUTPUT   Output path for the generated image
  -s, --random_seed RANDOM_SEED
                        Random seed for image generation
  --lora_dir LORA_DIR   Path to directory containing LoRA files
  --wtype WTYPE         Weight type (f32, f16, q4_0, q4_1, q5_0, q5_1, q8_0)
  --control_net_path CONTROL_NET_PATH
                        Path to control net model
  --control_image_path CONTROL_IMAGE_PATH
                        Path to image condition for Control Net
  --control_strength CONTROL_STRENGTH
                        Strength to apply Control Net

Example Command:

nexa run sd1-4

Run vision-language model

Run vision-language models on your local computer.

nexa run MODEL_PATH
usage: nexa run [-h] [-t TEMPERATURE] [-m MAX_NEW_TOKENS] [-k TOP_K] [-p TOP_P] [-sw [STOP_WORDS ...]] [-pf] [-st] model_path

positional arguments:
  model_path            Path or identifier for the model in Nexa Model Hub

options:
  -h, --help            show this help message and exit
  -pf, --profiling      Enable profiling logs for the inference process
  -st, --streamlit      Run the inference in Streamlit UI, can be used with -lp or -hf
  -lp, --local_path     Indicate that the model path provided is the local path, must be used with -mt
  -mt, --model_type     Indicate the model running type, must be used with -lp or -hf, choose from [NLP, COMPUTER_VISION, MULTIMODAL, AUDIO]
  -hf, --huggingface    Load model from Hugging Face Hub, must be used with -mt
  
VLM generation options:
  -t, --temperature TEMPERATURE
                        Temperature for sampling
  -m, --max_new_tokens MAX_NEW_TOKENS
                        Maximum number of new tokens to generate
  -k, --top_k TOP_K     Top-k sampling parameter
  -p, --top_p TOP_P     Top-p sampling parameter
  -sw, --stop_words [STOP_WORDS ...]
                        List of stop words for early stopping

Example Command:

nexa run nanollava

Run audio model

Run audio models on your local computer.

nexa run MODEL_PATH
usage: nexa run [-h] [-o OUTPUT_DIR] [-b BEAM_SIZE] [-l LANGUAGE] [--task TASK] [-t TEMPERATURE] [-c COMPUTE_TYPE] [-st] model_path

positional arguments:
  model_path            Path or identifier for the model in Nexa Model Hub

options:
  -h, --help            show this help message and exit
  -st, --streamlit      Run the inference in Streamlit UI, can be used with -lp or -hf
  -lp, --local_path     Indicate that the model path provided is the local path, must be used with -mt
  -mt, --model_type     Indicate the model running type, must be used with -lp or -hf, choose from [NLP, COMPUTER_VISION, MULTIMODAL, AUDIO]
  -hf, --huggingface    Load model from Hugging Face Hub, must be used with -mt
  
Automatic Speech Recognition options:
  -b, --beam_size BEAM_SIZE
                        Beam size to use for transcription
  -l, --language LANGUAGE
                        The language spoken in the audio. It should be a language code such as 'en' or 'fr'.
  --task TASK           Task to execute (transcribe or translate)
  -c, --compute_type COMPUTE_TYPE
                        Type to use for computation (e.g., float16, int8, int8_float16)

Example Command:

nexa run faster-whisper-tiny

Generate Embeddings

Generate Text Embeddings

nexa embed MODEL_PATH
usage: nexa embed [-h] [-lp] [-hf] [-n] [-nt] model_path prompt

positional arguments:
  model_path            Path or identifier for the model in Nexa Model Hub
  prompt                Prompt to generate embeddings

options:
  -h, --help            show this help message and exit
  -lp, --local_path     Indicate that the model path provided is the local path, must be used with -mt
  -hf, --huggingface    Load model from Hugging Face Hub, must be used with -mt
  -n, --normalize       Normalize the embeddings
  -nt, --no_truncate    Not truncate the embeddings

Example Command:

nexa embed mxbai "I love Nexa AI."
nexa embed nomic "I love Nexa AI." >> generated_embeddings.txt
nexa embed nomic-embed-text-v1.5:fp16 "I love Nexa AI." 
nexa embed sentence-transformers/all-MiniLM-L6-v2:gguf-fp16 "I love Nexa AI." >> generated_embeddings.txt

Start local server

start a local server using models on your local computer.

nexa server MODEL_PATH
usage: nexa server [-h] [--host HOST] [--port PORT] [--reload] [--nctx NCTX] model_path

positional arguments:
  model_path   Path or identifier for the model in S3

options:
  -h, --help   show this help message and exit
  -lp, --local_path     Indicate that the model path provided is the local path, must be used with -mt
  -mt, --model_type     Indicate the model running type, must be used with -lp or -hf, choose from [NLP, COMPUTER_VISION, MULTIMODAL, AUDIO]
  -hf, --huggingface    Load model from Hugging Face Hub, must be used with -mt
  --host HOST  Host to bind the server to
  --port PORT  Port to bind the server to
  --reload     Enable automatic reloading on code changes

Example Command:

nexa server llama2

Run Model Evaluation

Run evaluation using models on your local computer.

usage: nexa eval model_path [-h] [--tasks TASKS] [--limit LIMIT]

positional arguments:
  model_path            Path or identifier for the model in Nexa Model Hub

options:
  -h, --help            show this help message and exit
  --tasks TASKS         Tasks to evaluate, comma-separated
  --limit LIMIT         Limit the number of examples per task. If <1, limit is a percentage of the total number of examples.

Examples

nexa eval phi3 --tasks ifeval --limit 0.5

For more details, please refer to Local Server page

For model_path in nexa commands,

If you want to use models in Nexa Model Hub, it's better to follow the standard format as below to ensure correct model loading and execution:

[user_name]/[repo_name]:[tag_name](user's model)
[repo_name]:[tag_name](official model)

If you want to use models in Hugging Face, use Hugging Face repo_id as model_path

PreviousUpload Model NextPython Interface

Last updated 1 month ago