convert-to-rkllm / mapping.json
xiaoyao9184's picture
Synced repo using 'sync_with_huggingface' Github Action
eb0bbf9 verified
{
"RK3588": {
"rk3588-3npu-w8a8": {
"dataset": "./data_quant.json",
"qparams": null,
"optimization_level": 1,
"target_platform": "RK3588",
"num_npu_core": 3,
"quantized_dtype": "W8A8",
"quantized_algorithm": "normal",
"max_context": 4096
},
"rk3588-3npu-w8a8_g128": {
"dataset": "./data_quant.json",
"qparams": null,
"optimization_level": 1,
"target_platform": "RK3588",
"num_npu_core": 3,
"quantized_dtype": "W8A8_G128",
"quantized_algorithm": "normal",
"max_context": 4096
},
"rk3588-3npu-w8a8_g256": {
"dataset": "./data_quant.json",
"qparams": null,
"optimization_level": 1,
"target_platform": "RK3588",
"num_npu_core": 3,
"quantized_dtype": "W8A8_G256",
"quantized_algorithm": "normal",
"max_context": 4096
},
"rk3588-3npu-w8a8_g512": {
"dataset": "./data_quant.json",
"qparams": null,
"optimization_level": 1,
"target_platform": "RK3588",
"num_npu_core": 3,
"quantized_dtype": "W8A8_G512",
"quantized_algorithm": "normal",
"max_context": 4096
}
},
"RK3576": {
"rk3576-2npu-w8a8": {
"dataset": "./data_quant.json",
"qparams": null,
"optimization_level": 1,
"target_platform": "RK3576",
"num_npu_core": 2,
"quantized_dtype": "w8a8",
"quantized_algorithm": "normal",
"max_context": 4096
},
"rk3576-2npu-w4a16": {
"dataset": "./data_quant.json",
"qparams": null,
"optimization_level": 1,
"target_platform": "RK3576",
"num_npu_core": 2,
"quantized_dtype": "w4a16",
"quantized_algorithm": "grq",
"max_context": 4096
},
"rk3576-2npu-w4a16_g32": {
"dataset": "./data_quant.json",
"qparams": null,
"optimization_level": 1,
"target_platform": "RK3576",
"num_npu_core": 2,
"quantized_dtype": "w4a16_g32",
"quantized_algorithm": "grq",
"max_context": 4096
},
"rk3576-2npu-w4a16_g64": {
"dataset": "./data_quant.json",
"qparams": null,
"optimization_level": 1,
"target_platform": "RK3576",
"num_npu_core": 2,
"quantized_dtype": "w4a16_g64",
"quantized_algorithm": "grq",
"max_context": 4096
},
"rk3576-2npu-w4a16_g128": {
"dataset": "./data_quant.json",
"qparams": null,
"optimization_level": 1,
"target_platform": "RK3576",
"num_npu_core": 2,
"quantized_dtype": "w4a16_g128",
"quantized_algorithm": "grq",
"max_context": 4096
}
},
"RK3562": {
"rk3562-1npu-w8a8": {
"dataset": "./data_quant.json",
"qparams": null,
"optimization_level": 1,
"target_platform": "RK3562",
"quantized_dtype": "W8A8",
"quantized_algorithm": "normal",
"num_npu_core": 1,
"max_context": 4096
},
"rk3562-1npu-w4a16_g32": {
"dataset": "./data_quant.json",
"qparams": null,
"optimization_level": 1,
"target_platform": "RK3562",
"num_npu_core": 1,
"quantized_dtype": "W4A16_G32",
"quantized_algorithm": "grq",
"max_context": 4096
},
"rk3562-1npu-w4a16_g64": {
"dataset": "./data_quant.json",
"qparams": null,
"optimization_level": 1,
"target_platform": "RK3562",
"num_npu_core": 1,
"quantized_dtype": "W4A16_G64",
"quantized_algorithm": "grq",
"max_context": 4096
},
"rk3562-1npu-w4a16_g128": {
"dataset": "./data_quant.json",
"qparams": null,
"optimization_level": 1,
"target_platform": "RK3562",
"num_npu_core": 1,
"quantized_dtype": "W4A16_G128",
"quantized_algorithm": "grq",
"max_context": 4096
}
}
}