Qwen3.5-27B GGUF

• qwen3.5-27b-mix-13.4GiB.gguf

• token_embd/output = Q4_K
• attn_qkv = Q4_K
• full-attention attn_q/k/output = Q4_K
• ssm_alpha/ssm_beta = Q6_K
• attn_gate = Q4_0
• ssm_out = Q4_0
• ffn_down = Q4_0
• attention-block ffn_gate/up = Q4_K
• SSM-block ffn_gate/up = Q3_K

• quant size: 13678.55 MiB
• target: under 14 GiB

• GPU load test passed on 2x NVIDIA A40
• Fast MMLU-Pro sampled check against FP16 showed about -1.04 percentage points over 3 seeds (sample=64)

• /home/kasm-user/models/Qwen3.5-27B/qwen3.5-27b-mix-lt14gb.gguf