SFR-Embedding-2_R-4bit-NF4 / quantization_info.json

Add 4-bit NF4 mixed-precision quantized ST pipeline with detailed results-aware model card

1fb0d9b verified about 1 year ago

278 Bytes

{
  "quantization_method": "bitsandbytes_nf4",
  "base_model": "Salesforce/SFR-Embedding-2_R",
  "skip_patterns": [
    "embed_tokens",
    "layernorm",
    "norm",
    "layers.0.self_attn",
    "layers.1.self_attn"
  ],
  "size_mb": 3578.508056640625,
  "phase1_outliers": 31
}

	{
	"quantization_method": "bitsandbytes_nf4",
	"base_model": "Salesforce/SFR-Embedding-2_R",
	"skip_patterns": [
	"embed_tokens",
	"layernorm",
	"norm",
	"layers.0.self_attn",
	"layers.1.self_attn"
	],
	"size_mb": 3578.508056640625,
	"phase1_outliers": 31
	}