Without Internet Access

If you're deploying Compressa in a closed network without internet access, you need to download model weights from HuggingFace or through the Compressa platform in advance. You also need to download Docker images in advance.

Configuration File

If the model was not downloaded through the platform, you need to create a configuration file compressa-config.json in the folder with model weights

compressa-config.json
config.json
generation_config.json
.gitattributes
LICENSE
merges.txt
model-00001-of-00008.safetensors
model-00002-of-00008.safetensors
model-00003-of-00008.safetensors
model-00004-of-00008.safetensors
model-00005-of-00008.safetensors
model-00006-of-00008.safetensors
model-00007-of-00008.safetensors
model-00008-of-00008.safetensors
model.safetensors.index.json
README.md
tokenizer_config.json
tokenizer.json
vocab.json

{
    "model_name": "Qwen/Qwen3-14B",
    "torch_dtype": "torch.float16",
    "device_map": "auto",
    "trust_remote_code": false,
    "backends": null,
    "config_class": "HFConfig"
}

Setting Environment Variable

Set the RESOURCES_PATH environment variable to the data/compressa folder from the archive.

export RESOURCES_PATH=./data/compressa

Start Compressa as usual.

The model will be available from the first launch without the need for downloading.