test,model,architecture,quant,file_gib,engine,config,metric,value,unit,runs
arch_compare,Qwen3.6-35B-A3B,moe_3b_active,Q4_K_M,20.22,llama.cpp,default,pp4096,671.40,tok/s,3
arch_compare,Qwen3.6-35B-A3B,moe_3b_active,Q4_K_M,20.22,llama.cpp,default,tg128,58.56,tok/s,3
arch_compare,Qwen3.8-27B,dense,IQ3_S,11.20,llama.cpp,default,pp4096,106.78,tok/s,3
arch_compare,Qwen3.8-27B,dense,IQ3_S,11.20,llama.cpp,default,tg128,10.82,tok/s,3
quant_sweep,Qwen3.8-27B,dense,IQ4_XS,13.32,ollama,32k ctx q8_0 kv,generation,10.5,tok/s,2
quant_sweep,Qwen3.8-27B,dense,IQ2_S,7.82,ollama,32k ctx q8_0 kv,generation,11.9,tok/s,2
mtp_cost,Qwen3.6-35B-A3B,moe_3b_active,Q4_K_M,20.22,ollama,32k ctx q8_0 kv draft-mtp n=2,generation,34.6,tok/s,3
mtp_cost,Qwen3.6-35B-A3B,moe_3b_active,Q4_K_M,20.22,ollama,32k ctx q8_0 kv no draft,generation,46.2,tok/s,3
mtp_depth,Qwen3.6-35B-A3B,moe_3b_active,Q4_K_M,20.22,ollama,draft-mtp n=2,generation,42.0,tok/s,2
mtp_depth,Qwen3.6-35B-A3B,moe_3b_active,Q4_K_M,20.22,ollama,draft-mtp n=4,generation,29.4,tok/s,2
mtp_depth,Qwen3.6-35B-A3B,moe_3b_active,Q4_K_M,20.22,ollama,draft-mtp n=6,generation,20.1,tok/s,2
mtp_depth,Qwen3.6-35B-A3B,moe_3b_active,Q4_K_M,20.22,ollama,draft-mtp n=8,generation,17.1,tok/s,2
kv_cache,Qwen3.6-35B-A3B,moe_3b_active,Q4_K_M,20.22,llama.cpp,f16 kv,tg128,58.55,tok/s,3
kv_cache,Qwen3.6-35B-A3B,moe_3b_active,Q4_K_M,20.22,llama.cpp,q8_0 kv,tg128,57.25,tok/s,3
engine_compare,Qwen3.6-35B-A3B,moe_3b_active,4bit,20.22,mlx-lm,alone on gpu,prefill,296,tok/s,1
engine_compare,Qwen3.6-35B-A3B,moe_3b_active,4bit,20.22,mlx-lm,alone on gpu,generation,86.9,tok/s,1
engine_compare,Qwen3.6-35B-A3B,moe_3b_active,Q4_K_M,20.22,llama.cpp,alone on gpu,prefill,685,tok/s,3
engine_compare,Qwen3.6-35B-A3B,moe_3b_active,Q4_K_M,20.22,llama.cpp,alone on gpu,generation,57.8,tok/s,3
sample_length,Qwen3.6-35B-A3B,moe_3b_active,Q4_K_M,20.22,ollama,24-token sample,generation,56.5,tok/s,3
sample_length,Qwen3.6-35B-A3B,moe_3b_active,Q4_K_M,20.22,ollama,700-token sample,generation,41.0,tok/s,3
quant_sweep,Qwen3.8-27B,dense,4bit,15.34,oMLX,"8k ctx, MLX affine (not GGUF), balanced guard, ssd cache on",pp8192,84,tok/s,5
quant_sweep,Qwen3.8-27B,dense,4bit,15.34,oMLX,"8k ctx, MLX affine (not GGUF), balanced guard, ssd cache on",generation,13.7,tok/s,5
dflash2,Qwen3.8-27B,dense,IQ3_S,11.20,llama.cpp,"code prompt, 256 tok, no spec decoding",generation,10.53,tok/s,2
dflash2,Qwen3.8-27B,dense,IQ3_S,11.20,llama.cpp,"code prompt, 256 tok, draft-dflash n-max 7",generation,10.18,tok/s,2
dflash2,Qwen3.8-27B,dense,IQ3_S,11.20,llama.cpp,"math prompt, 256 tok, no spec decoding",generation,10.30,tok/s,2
dflash2,Qwen3.8-27B,dense,IQ3_S,11.20,llama.cpp,"math prompt, 256 tok, draft-dflash n-max 7",generation,9.81,tok/s,2
dflash2,Qwen3.8-27B,dense,IQ3_S,11.20,llama.cpp,"prose prompt, 256 tok, no spec decoding",generation,10.29,tok/s,2
dflash2,Qwen3.8-27B,dense,IQ3_S,11.20,llama.cpp,"prose prompt, 256 tok, draft-dflash n-max 7",generation,6.13,tok/s,2
dflash2,Qwen3.8-27B,dense,IQ3_S,11.20,llama.cpp,"code prompt, draft-dflash n-max 7",draft_acceptance,0.76,ratio,2
dflash2,Qwen3.8-27B,dense,IQ3_S,11.20,llama.cpp,"math prompt, draft-dflash n-max 7",draft_acceptance,0.72,ratio,2
dflash2,Qwen3.8-27B,dense,IQ3_S,11.20,llama.cpp,"prose prompt, draft-dflash n-max 7",draft_acceptance,0.35,ratio,2
dflash2,Qwen3.6-35B-A3B,moe_3b_active,IQ3_S,12.74,llama.cpp,"code prompt 256 tok no spec",generation,55.28,tok/s,2
dflash2,Qwen3.6-35B-A3B,moe_3b_active,IQ3_S,12.74,llama.cpp,"code prompt 256 tok draft-dflash n-max 7",generation,49.26,tok/s,2
dflash2,Qwen3.6-35B-A3B,moe_3b_active,IQ3_S,12.74,llama.cpp,"math prompt 256 tok no spec",generation,52.09,tok/s,2
dflash2,Qwen3.6-35B-A3B,moe_3b_active,IQ3_S,12.74,llama.cpp,"math prompt 256 tok draft-dflash n-max 7",generation,48.50,tok/s,2
dflash2,Qwen3.6-35B-A3B,moe_3b_active,IQ3_S,12.74,llama.cpp,"prose prompt 256 tok no spec",generation,51.02,tok/s,2
dflash2,Qwen3.6-35B-A3B,moe_3b_active,IQ3_S,12.74,llama.cpp,"prose prompt 256 tok draft-dflash n-max 7",generation,30.79,tok/s,2
dflash2,Qwen3.6-35B-A3B,moe_3b_active,IQ3_S,12.74,llama.cpp,"code prompt draft-dflash n-max 7",draft_acceptance,0.60,ratio,2
dflash2,Qwen3.6-35B-A3B,moe_3b_active,IQ3_S,12.74,llama.cpp,"math prompt draft-dflash n-max 7",draft_acceptance,0.59,ratio,2
dflash2,Qwen3.6-35B-A3B,moe_3b_active,IQ3_S,12.74,llama.cpp,"prose prompt draft-dflash n-max 7",draft_acceptance,0.28,ratio,2