Sakura-EmbeddingGemma-2-AutoRound / quantization_config.json

v2: re-quantized W4A16 (group 64, 200 iters, 234 real retrieval calibration texts); modest gains in BF16 fidelity and BEIR; README re-measured, build manifest added. v1 stays in git history.

3f8b39e verified 2 days ago

Raw History Blame Contribute Delete

290 Bytes

	{
	"bits": 4,
	"data_type": "int",
	"group_size": 64,
	"sym": true,
	"low_gpu_mem_usage": true,
	"autoround_version": "0.16.0",
	"block_name_to_quantize": "language_model.layers",
	"quant_method": "auto-round",
	"packing_format": "auto_round:auto_gptq",
	"iters": 200
	}