Luigi/x-asr-zh-en-streaming-zipformer2-gguf

Nano benchmark: 4-engine cross-comparison + thread scaling (sherpa-onnx CUDA not runnable; RS CUDA+FP16 reaches sherpa-CPU parity)

Add measured on-device Jetson Nano (sm_53) benchmark: RS_GEMM_FP16 1.75x, token-exact

Update model card: add iq4_xs, per-weight accuracy table, document excluded sub-3-bit sweep

Add 960ms iq4_xs (lossless, 87MB)

Document q3_k (imatrix) near-lossless 69MB option

Add q3_k (imatrix, near-lossless, 69MB) + imatrix .dat calibration files

Document q8_0/q4_k as recommended formats + quant sweep guidance

Add Q8_0 variants (lossless: token-exact with f16, ~half size, ~1.2x faster on CPU)

Add all 4 streaming chunk variants (f16 + Q4_K) + model card

initial commit