ERNIE-Image-INT8 / precision_manifest.json
ixim's picture
Add files using upload-large-folder tool
b90c5ef verified
{
"device": "cuda",
"base_model_dir": "Baidu/ERNIE-Image",
"base_model_hub_id": "Baidu/ERNIE-Image",
"base_model_relation": "quantized",
"components": [
{
"name": "transformer",
"backend": "quanto",
"precision": "int8",
"enabled": true,
"skipped_modules": [
"norm",
"layer_norm",
"ln",
"embed_tokens",
"lm_head",
"proj_out"
]
},
{
"name": "text_encoder",
"backend": "none",
"precision": "bfloat16",
"enabled": false,
"skipped_modules": []
},
{
"name": "pe",
"backend": "none",
"precision": "bfloat16",
"enabled": false,
"skipped_modules": []
}
]
}