This is a GPTQ quantized version of kimiko-llongma-2-13B-16k

To run this model, make sure compress_pos_emb is set to 4 to apply proper rope scaling parameters. The max_ctx_len is 16384.

Branches:

• main: 4 bits, groupsize 128, act order false
• 4bit-128g-actorder: 4 bits, groupsize 128, act order true
• 4bit-32g-actorder: 4 bits, groupsize 32, act order true