Datasets:

pbhappliedsystems
/

quant_eval_paired_degradation_statistics

run_id stringclasses 6 values	model_id stringclasses 4 values	baseline_quant_type stringclasses 2 values	quantized_quant_type stringclasses 4 values	family stringclasses 8 values	runner_a stringclasses 2 values	runner_b stringclasses 2 values	paired_n int64 200 200	both_pass int64 69 200	both_fail int64 0 117	runner_a_only_pass int64 0 56	runner_b_only_pass int64 0 39	disagreements int64 0 56	disagreement_rate float64 0 0.28	pass_rate_delta_runner_b_minus_runner_a float64 -0.28 0.19	delta_ci_low float64 -0.36 0.1	delta_ci_high float64 -0.19 0.26	delta_ci_level float64 0.95 0.95	delta_ci_method stringclasses 1 value	mcnemar_p_value float64 0 1	mcnemar_discordant_pairs int64 0 56	mcnemar_alternative stringclasses 1 value	discriminating bool 2 classes	cluster_adjusted_method stringclasses 1 value	cluster_adjusted_effective_paired_n int64 167 200	cluster_adjusted_delta float64 -0.28 0.19	cluster_adjusted_delta_ci_low float64 -0.36 0.09	cluster_adjusted_delta_ci_high float64 -0.19 0.27
Mistral_Nemo_Instruct_2407_20260814_030505	mistralai/Mistral-Nemo-Instruct-2407	gguf_f16	gguf_q4_k_m	json	full_weight_llama_cpp	quantized_llama_cpp	200	81	113	2	4	6	0.03	0.01	-0.034783	0.054293	0.95	paired_multinomial_bonferroni_wilson	0.6875	6	two-sided	true	semantic-template clusters; mixed-direction clusters conservatively count as baseline-only pass	200	0.01	-0.034783	0.054293
Mistral_Nemo_Instruct_2407_20260814_030505	mistralai/Mistral-Nemo-Instruct-2407	gguf_f16	gguf_q4_k_m	json_multistep	full_weight_llama_cpp	quantized_llama_cpp	200	95	76	6	23	29	0.145	0.085	0.003014	0.16282	0.95	paired_multinomial_bonferroni_wilson	0.002316	29	two-sided	true	semantic-template clusters; mixed-direction clusters conservatively count as baseline-only pass	167	0.047904	-0.036963	0.129973
Mistral_Nemo_Instruct_2407_20260814_030505	mistralai/Mistral-Nemo-Instruct-2407	gguf_f16	gguf_q4_k_m	mcq	full_weight_llama_cpp	quantized_llama_cpp	200	167	14	17	2	19	0.095	-0.075	-0.137652	-0.008672	0.95	paired_multinomial_bonferroni_wilson	0.000729	19	two-sided	true	semantic-template clusters; mixed-direction clusters conservatively count as baseline-only pass	200	-0.075	-0.137652	-0.008672
Mistral_Nemo_Instruct_2407_20260814_030505	mistralai/Mistral-Nemo-Instruct-2407	gguf_f16	gguf_q4_k_m	mixed_brief_json	full_weight_llama_cpp	quantized_llama_cpp	200	144	0	56	0	56	0.28	-0.28	-0.355882	-0.190395	0.95	paired_multinomial_bonferroni_wilson	0	56	two-sided	true	semantic-template clusters; mixed-direction clusters conservatively count as baseline-only pass	200	-0.28	-0.355882	-0.190395
Mistral_Nemo_Instruct_2407_20260814_030505	mistralai/Mistral-Nemo-Instruct-2407	gguf_f16	gguf_q4_k_m	stateful_followup	full_weight_llama_cpp	quantized_llama_cpp	200	166	0	34	0	34	0.17	-0.17	-0.23744	-0.094228	0.95	paired_multinomial_bonferroni_wilson	0	34	two-sided	true	semantic-template clusters; mixed-direction clusters conservatively count as baseline-only pass	200	-0.17	-0.23744	-0.094228
Mistral_Nemo_Instruct_2407_20260814_030505	mistralai/Mistral-Nemo-Instruct-2407	gguf_f16	gguf_q4_k_m	toolcall	full_weight_llama_cpp	quantized_llama_cpp	200	147	0	39	14	53	0.265	-0.125	-0.225712	-0.018162	0.95	paired_multinomial_bonferroni_wilson	0.000802	53	two-sided	true	semantic-template clusters; mixed-direction clusters conservatively count as baseline-only pass	200	-0.125	-0.225712	-0.018162
Mistral_Nemo_Instruct_2407_20260814_030505	mistralai/Mistral-Nemo-Instruct-2407	gguf_f16	gguf_q4_k_m	toolcall_only	full_weight_llama_cpp	quantized_llama_cpp	200	157	0	43	0	43	0.215	-0.215	-0.286671	-0.132793	0.95	paired_multinomial_bonferroni_wilson	0	43	two-sided	true	semantic-template clusters; mixed-direction clusters conservatively count as baseline-only pass	200	-0.215	-0.286671	-0.132793
Mistral_Nemo_Instruct_2407_20260814_030505	mistralai/Mistral-Nemo-Instruct-2407	gguf_f16	gguf_q4_k_m	fuzz	full_weight_llama_cpp	quantized_llama_cpp	200	103	92	2	3	5	0.025	0.005	-0.037223	0.046978	0.95	paired_multinomial_bonferroni_wilson	1	5	two-sided	true	semantic-template clusters; mixed-direction clusters conservatively count as baseline-only pass	200	0.005	-0.037223	0.046978
Mistral_Nemo_Instruct_2407_20260815_113254	mistralai/Mistral-Nemo-Instruct-2407	gguf_f16	gguf_q5_k_m	json	full_weight_llama_cpp	quantized_llama_cpp	200	83	117	0	0	0	0	0	-0.024504	0.024504	0.95	paired_multinomial_bonferroni_wilson	1	0	two-sided	false	semantic-template clusters; mixed-direction clusters conservatively count as baseline-only pass	200	0	-0.024504	0.024504
Mistral_Nemo_Instruct_2407_20260815_113254	mistralai/Mistral-Nemo-Instruct-2407	gguf_f16	gguf_q5_k_m	json_multistep	full_weight_llama_cpp	quantized_llama_cpp	200	101	98	0	1	1	0.005	0.005	-0.023777	0.033532	0.95	paired_multinomial_bonferroni_wilson	1	1	two-sided	true	semantic-template clusters; mixed-direction clusters conservatively count as baseline-only pass	167	0.005988	-0.028333	0.03996
Mistral_Nemo_Instruct_2407_20260815_113254	mistralai/Mistral-Nemo-Instruct-2407	gguf_f16	gguf_q5_k_m	mcq	full_weight_llama_cpp	quantized_llama_cpp	200	181	15	3	1	4	0.02	-0.01	-0.048591	0.029081	0.95	paired_multinomial_bonferroni_wilson	0.625	4	two-sided	true	semantic-template clusters; mixed-direction clusters conservatively count as baseline-only pass	200	-0.01	-0.048591	0.029081
Mistral_Nemo_Instruct_2407_20260815_113254	mistralai/Mistral-Nemo-Instruct-2407	gguf_f16	gguf_q5_k_m	mixed_brief_json	full_weight_llama_cpp	quantized_llama_cpp	200	200	0	0	0	0	0	0	-0.024504	0.024504	0.95	paired_multinomial_bonferroni_wilson	1	0	two-sided	false	semantic-template clusters; mixed-direction clusters conservatively count as baseline-only pass	200	0	-0.024504	0.024504
Mistral_Nemo_Instruct_2407_20260815_113254	mistralai/Mistral-Nemo-Instruct-2407	gguf_f16	gguf_q5_k_m	stateful_followup	full_weight_llama_cpp	quantized_llama_cpp	200	200	0	0	0	0	0	0	-0.024504	0.024504	0.95	paired_multinomial_bonferroni_wilson	1	0	two-sided	false	semantic-template clusters; mixed-direction clusters conservatively count as baseline-only pass	200	0	-0.024504	0.024504
Mistral_Nemo_Instruct_2407_20260815_113254	mistralai/Mistral-Nemo-Instruct-2407	gguf_f16	gguf_q5_k_m	toolcall	full_weight_llama_cpp	quantized_llama_cpp	200	184	14	2	0	2	0.01	-0.01	-0.041673	0.022163	0.95	paired_multinomial_bonferroni_wilson	0.5	2	two-sided	true	semantic-template clusters; mixed-direction clusters conservatively count as baseline-only pass	200	-0.01	-0.041673	0.022163
Mistral_Nemo_Instruct_2407_20260815_113254	mistralai/Mistral-Nemo-Instruct-2407	gguf_f16	gguf_q5_k_m	toolcall_only	full_weight_llama_cpp	quantized_llama_cpp	200	199	0	1	0	1	0.005	-0.005	-0.033532	0.023777	0.95	paired_multinomial_bonferroni_wilson	1	1	two-sided	true	semantic-template clusters; mixed-direction clusters conservatively count as baseline-only pass	200	-0.005	-0.033532	0.023777
Mistral_Nemo_Instruct_2407_20260815_113254	mistralai/Mistral-Nemo-Instruct-2407	gguf_f16	gguf_q5_k_m	fuzz	full_weight_llama_cpp	quantized_llama_cpp	200	103	92	2	3	5	0.025	0.005	-0.037223	0.046978	0.95	paired_multinomial_bonferroni_wilson	1	5	two-sided	true	semantic-template clusters; mixed-direction clusters conservatively count as baseline-only pass	200	0.005	-0.037223	0.046978
Mistral_Nemo_Instruct_2407_20260816_084553	mistralai/Mistral-Nemo-Instruct-2407	gguf_f16	gguf_q8_0	json	full_weight_llama_cpp	quantized_llama_cpp	200	83	117	0	0	0	0	0	-0.024504	0.024504	0.95	paired_multinomial_bonferroni_wilson	1	0	two-sided	false	semantic-template clusters; mixed-direction clusters conservatively count as baseline-only pass	200	0	-0.024504	0.024504
Mistral_Nemo_Instruct_2407_20260816_084553	mistralai/Mistral-Nemo-Instruct-2407	gguf_f16	gguf_q8_0	json_multistep	full_weight_llama_cpp	quantized_llama_cpp	200	101	99	0	0	0	0	0	-0.024504	0.024504	0.95	paired_multinomial_bonferroni_wilson	1	0	two-sided	false	semantic-template clusters; mixed-direction clusters conservatively count as baseline-only pass	167	0	-0.029205	0.029205
Mistral_Nemo_Instruct_2407_20260816_084553	mistralai/Mistral-Nemo-Instruct-2407	gguf_f16	gguf_q8_0	mcq	full_weight_llama_cpp	quantized_llama_cpp	200	183	15	1	1	2	0.01	0	-0.032804	0.032804	0.95	paired_multinomial_bonferroni_wilson	1	2	two-sided	true	semantic-template clusters; mixed-direction clusters conservatively count as baseline-only pass	200	0	-0.032804	0.032804
Mistral_Nemo_Instruct_2407_20260816_084553	mistralai/Mistral-Nemo-Instruct-2407	gguf_f16	gguf_q8_0	mixed_brief_json	full_weight_llama_cpp	quantized_llama_cpp	200	200	0	0	0	0	0	0	-0.024504	0.024504	0.95	paired_multinomial_bonferroni_wilson	1	0	two-sided	false	semantic-template clusters; mixed-direction clusters conservatively count as baseline-only pass	200	0	-0.024504	0.024504
Mistral_Nemo_Instruct_2407_20260816_084553	mistralai/Mistral-Nemo-Instruct-2407	gguf_f16	gguf_q8_0	stateful_followup	full_weight_llama_cpp	quantized_llama_cpp	200	200	0	0	0	0	0	0	-0.024504	0.024504	0.95	paired_multinomial_bonferroni_wilson	1	0	two-sided	false	semantic-template clusters; mixed-direction clusters conservatively count as baseline-only pass	200	0	-0.024504	0.024504
Mistral_Nemo_Instruct_2407_20260816_084553	mistralai/Mistral-Nemo-Instruct-2407	gguf_f16	gguf_q8_0	toolcall	full_weight_llama_cpp	quantized_llama_cpp	200	186	14	0	0	0	0	0	-0.024504	0.024504	0.95	paired_multinomial_bonferroni_wilson	1	0	two-sided	false	semantic-template clusters; mixed-direction clusters conservatively count as baseline-only pass	200	0	-0.024504	0.024504
Mistral_Nemo_Instruct_2407_20260816_084553	mistralai/Mistral-Nemo-Instruct-2407	gguf_f16	gguf_q8_0	toolcall_only	full_weight_llama_cpp	quantized_llama_cpp	200	199	0	1	0	1	0.005	-0.005	-0.033532	0.023777	0.95	paired_multinomial_bonferroni_wilson	1	1	two-sided	true	semantic-template clusters; mixed-direction clusters conservatively count as baseline-only pass	200	-0.005	-0.033532	0.023777
Mistral_Nemo_Instruct_2407_20260816_084553	mistralai/Mistral-Nemo-Instruct-2407	gguf_f16	gguf_q8_0	fuzz	full_weight_llama_cpp	quantized_llama_cpp	200	104	93	1	2	3	0.015	0.005	-0.031191	0.040946	0.95	paired_multinomial_bonferroni_wilson	1	3	two-sided	true	semantic-template clusters; mixed-direction clusters conservatively count as baseline-only pass	200	0.005	-0.031191	0.040946
Qwen2.5_14B_Instruct_1M_20260815_220633	Qwen/Qwen2.5-14B-Instruct-1M	modal_f16	modal_q4_k_m	json	full_weight_modal_llama_cpp	quantized_modal_llama_cpp	200	98	71	16	15	31	0.155	-0.005	-0.0911	0.081345	0.95	paired_multinomial_bonferroni_wilson	1	31	two-sided	true	semantic-template clusters; mixed-direction clusters conservatively count as baseline-only pass	200	-0.005	-0.0911	0.081345
Qwen2.5_14B_Instruct_1M_20260815_220633	Qwen/Qwen2.5-14B-Instruct-1M	modal_f16	modal_q4_k_m	json_multistep	full_weight_modal_llama_cpp	quantized_modal_llama_cpp	200	105	81	2	12	14	0.07	0.05	-0.009599	0.107148	0.95	paired_multinomial_bonferroni_wilson	0.012939	14	two-sided	true	semantic-template clusters; mixed-direction clusters conservatively count as baseline-only pass	167	0.05988	-0.011163	0.127426
Qwen2.5_14B_Instruct_1M_20260815_220633	Qwen/Qwen2.5-14B-Instruct-1M	modal_f16	modal_q4_k_m	mcq	full_weight_modal_llama_cpp	quantized_modal_llama_cpp	200	162	4	32	2	34	0.17	-0.15	-0.22398	-0.068669	0.95	paired_multinomial_bonferroni_wilson	0	34	two-sided	true	semantic-template clusters; mixed-direction clusters conservatively count as baseline-only pass	200	-0.15	-0.22398	-0.068669
Qwen2.5_14B_Instruct_1M_20260815_220633	Qwen/Qwen2.5-14B-Instruct-1M	modal_f16	modal_q4_k_m	mixed_brief_json	full_weight_modal_llama_cpp	quantized_modal_llama_cpp	200	147	39	9	5	14	0.07	-0.02	-0.080892	0.041872	0.95	paired_multinomial_bonferroni_wilson	0.42395	14	two-sided	true	semantic-template clusters; mixed-direction clusters conservatively count as baseline-only pass	200	-0.02	-0.080892	0.041872
Qwen2.5_14B_Instruct_1M_20260815_220633	Qwen/Qwen2.5-14B-Instruct-1M	modal_f16	modal_q4_k_m	stateful_followup	full_weight_modal_llama_cpp	quantized_modal_llama_cpp	200	200	0	0	0	0	0	0	-0.024504	0.024504	0.95	paired_multinomial_bonferroni_wilson	1	0	two-sided	false	semantic-template clusters; mixed-direction clusters conservatively count as baseline-only pass	200	0	-0.024504	0.024504
Qwen2.5_14B_Instruct_1M_20260815_220633	Qwen/Qwen2.5-14B-Instruct-1M	modal_f16	modal_q4_k_m	toolcall	full_weight_modal_llama_cpp	quantized_modal_llama_cpp	200	190	2	0	8	8	0.04	0.04	-0.005912	0.083952	0.95	paired_multinomial_bonferroni_wilson	0.007813	8	two-sided	true	semantic-template clusters; mixed-direction clusters conservatively count as baseline-only pass	200	0.04	-0.005912	0.083952
Qwen2.5_14B_Instruct_1M_20260815_220633	Qwen/Qwen2.5-14B-Instruct-1M	modal_f16	modal_q4_k_m	toolcall_only	full_weight_modal_llama_cpp	quantized_modal_llama_cpp	200	175	19	1	5	6	0.03	0.02	-0.023962	0.062982	0.95	paired_multinomial_bonferroni_wilson	0.21875	6	two-sided	true	semantic-template clusters; mixed-direction clusters conservatively count as baseline-only pass	200	0.02	-0.023962	0.062982
Qwen2.5_14B_Instruct_1M_20260815_220633	Qwen/Qwen2.5-14B-Instruct-1M	modal_f16	modal_q4_k_m	fuzz	full_weight_modal_llama_cpp	quantized_modal_llama_cpp	200	121	53	6	20	26	0.13	0.07	-0.008769	0.145339	0.95	paired_multinomial_bonferroni_wilson	0.009355	26	two-sided	true	semantic-template clusters; mixed-direction clusters conservatively count as baseline-only pass	200	0.07	-0.008769	0.145339
Qwen2.5_32B_Instruct_20260815_081051	Qwen/Qwen2.5-32B-Instruct	modal_f16	modal_q4_k_m	json	full_weight_modal_llama_cpp	quantized_modal_llama_cpp	200	114	67	10	9	19	0.095	-0.005	-0.075044	0.065289	0.95	paired_multinomial_bonferroni_wilson	1	19	two-sided	true	semantic-template clusters; mixed-direction clusters conservatively count as baseline-only pass	200	-0.005	-0.075044	0.065289
Qwen2.5_32B_Instruct_20260815_081051	Qwen/Qwen2.5-32B-Instruct	modal_f16	modal_q4_k_m	json_multistep	full_weight_modal_llama_cpp	quantized_modal_llama_cpp	200	109	69	11	11	22	0.11	0	-0.074632	0.074632	0.95	paired_multinomial_bonferroni_wilson	1	22	two-sided	true	semantic-template clusters; mixed-direction clusters conservatively count as baseline-only pass	167	0.011976	-0.069721	0.092973
Qwen2.5_32B_Instruct_20260815_081051	Qwen/Qwen2.5-32B-Instruct	modal_f16	modal_q4_k_m	mcq	full_weight_modal_llama_cpp	quantized_modal_llama_cpp	200	200	0	0	0	0	0	0	-0.024504	0.024504	0.95	paired_multinomial_bonferroni_wilson	1	0	two-sided	false	semantic-template clusters; mixed-direction clusters conservatively count as baseline-only pass	200	0	-0.024504	0.024504
Qwen2.5_32B_Instruct_20260815_081051	Qwen/Qwen2.5-32B-Instruct	modal_f16	modal_q4_k_m	mixed_brief_json	full_weight_modal_llama_cpp	quantized_modal_llama_cpp	200	200	0	0	0	0	0	0	-0.024504	0.024504	0.95	paired_multinomial_bonferroni_wilson	1	0	two-sided	false	semantic-template clusters; mixed-direction clusters conservatively count as baseline-only pass	200	0	-0.024504	0.024504
Qwen2.5_32B_Instruct_20260815_081051	Qwen/Qwen2.5-32B-Instruct	modal_f16	modal_q4_k_m	stateful_followup	full_weight_modal_llama_cpp	quantized_modal_llama_cpp	200	200	0	0	0	0	0	0	-0.024504	0.024504	0.95	paired_multinomial_bonferroni_wilson	1	0	two-sided	false	semantic-template clusters; mixed-direction clusters conservatively count as baseline-only pass	200	0	-0.024504	0.024504
Qwen2.5_32B_Instruct_20260815_081051	Qwen/Qwen2.5-32B-Instruct	modal_f16	modal_q4_k_m	toolcall	full_weight_modal_llama_cpp	quantized_modal_llama_cpp	200	189	9	0	2	2	0.01	0.01	-0.022163	0.041673	0.95	paired_multinomial_bonferroni_wilson	0.5	2	two-sided	true	semantic-template clusters; mixed-direction clusters conservatively count as baseline-only pass	200	0.01	-0.022163	0.041673
Qwen2.5_32B_Instruct_20260815_081051	Qwen/Qwen2.5-32B-Instruct	modal_f16	modal_q4_k_m	toolcall_only	full_weight_modal_llama_cpp	quantized_modal_llama_cpp	200	197	2	1	0	1	0.005	-0.005	-0.033532	0.023777	0.95	paired_multinomial_bonferroni_wilson	1	1	two-sided	true	semantic-template clusters; mixed-direction clusters conservatively count as baseline-only pass	200	-0.005	-0.033532	0.023777
Qwen2.5_32B_Instruct_20260815_081051	Qwen/Qwen2.5-32B-Instruct	modal_f16	modal_q4_k_m	fuzz	full_weight_modal_llama_cpp	quantized_modal_llama_cpp	200	118	65	8	9	17	0.085	0.005	-0.062116	0.071871	0.95	paired_multinomial_bonferroni_wilson	1	17	two-sided	true	semantic-template clusters; mixed-direction clusters conservatively count as baseline-only pass	200	0.005	-0.062116	0.071871
Qwen2.5_7B_Instruct_20260814_234822	Qwen/Qwen2.5-7B-Instruct	gguf_f16	gguf_q4_k_m	json	full_weight_llama_cpp	quantized_llama_cpp	200	72	111	13	4	17	0.085	-0.045	-0.108805	0.02101	0.95	paired_multinomial_bonferroni_wilson	0.049042	17	two-sided	true	semantic-template clusters; mixed-direction clusters conservatively count as baseline-only pass	200	-0.045	-0.108805	0.02101
Qwen2.5_7B_Instruct_20260814_234822	Qwen/Qwen2.5-7B-Instruct	gguf_f16	gguf_q4_k_m	json_multistep	full_weight_llama_cpp	quantized_llama_cpp	200	69	90	2	39	41	0.205	0.185	0.098332	0.262602	0.95	paired_multinomial_bonferroni_wilson	0	41	two-sided	true	semantic-template clusters; mixed-direction clusters conservatively count as baseline-only pass	167	0.185629	0.088164	0.272251
Qwen2.5_7B_Instruct_20260814_234822	Qwen/Qwen2.5-7B-Instruct	gguf_f16	gguf_q4_k_m	mcq	full_weight_llama_cpp	quantized_llama_cpp	200	166	30	1	3	4	0.02	0.01	-0.029081	0.048591	0.95	paired_multinomial_bonferroni_wilson	0.625	4	two-sided	true	semantic-template clusters; mixed-direction clusters conservatively count as baseline-only pass	200	0.01	-0.029081	0.048591
Qwen2.5_7B_Instruct_20260814_234822	Qwen/Qwen2.5-7B-Instruct	gguf_f16	gguf_q4_k_m	mixed_brief_json	full_weight_llama_cpp	quantized_llama_cpp	200	185	0	15	0	15	0.075	-0.075	-0.12794	-0.018385	0.95	paired_multinomial_bonferroni_wilson	0.000061	15	two-sided	true	semantic-template clusters; mixed-direction clusters conservatively count as baseline-only pass	200	-0.075	-0.12794	-0.018385
Qwen2.5_7B_Instruct_20260814_234822	Qwen/Qwen2.5-7B-Instruct	gguf_f16	gguf_q4_k_m	stateful_followup	full_weight_llama_cpp	quantized_llama_cpp	200	150	26	0	24	24	0.12	0.12	0.053094	0.181025	0.95	paired_multinomial_bonferroni_wilson	0	24	two-sided	true	semantic-template clusters; mixed-direction clusters conservatively count as baseline-only pass	200	0.12	0.053094	0.181025
Qwen2.5_7B_Instruct_20260814_234822	Qwen/Qwen2.5-7B-Instruct	gguf_f16	gguf_q4_k_m	toolcall	full_weight_llama_cpp	quantized_llama_cpp	200	128	41	14	17	31	0.155	0.015	-0.071498	0.100763	0.95	paired_multinomial_bonferroni_wilson	0.7201	31	two-sided	true	semantic-template clusters; mixed-direction clusters conservatively count as baseline-only pass	200	0.015	-0.071498	0.100763
Qwen2.5_7B_Instruct_20260814_234822	Qwen/Qwen2.5-7B-Instruct	gguf_f16	gguf_q4_k_m	toolcall_only	full_weight_llama_cpp	quantized_llama_cpp	200	160	35	1	4	5	0.025	0.015	-0.026642	0.055907	0.95	paired_multinomial_bonferroni_wilson	0.375	5	two-sided	true	semantic-template clusters; mixed-direction clusters conservatively count as baseline-only pass	200	0.015	-0.026642	0.055907
Qwen2.5_7B_Instruct_20260814_234822	Qwen/Qwen2.5-7B-Instruct	gguf_f16	gguf_q4_k_m	fuzz	full_weight_llama_cpp	quantized_llama_cpp	200	93	88	11	8	19	0.095	-0.015	-0.084629	0.055364	0.95	paired_multinomial_bonferroni_wilson	0.647606	19	two-sided	true	semantic-template clusters; mixed-direction clusters conservatively count as baseline-only pass	200	-0.015	-0.084629	0.055364

quant_eval — Paired degradation statistics

One row per run per task family: the paired pass-rate difference with a 95% confidence interval, the two-sided exact McNemar test, the full discordance breakdown, and a semantic-cluster-adjusted delta and interval.

Part of the quant_eval public corpus: a per-case behavioral evaluation of full-weight and quantized large language models across eight agent-relevant task families, with paired statistical testing.

Cite this dataset: 10.5281/zenodo.22010557 — concept DOI, always resolves to the latest version. This exact deposit: 10.5281/zenodo.22010558 — version DOI, frozen. Cite this one where reported numbers must stay verifiable against the object referenced.

What this file contains

File	Rows	Columns
`quant_eval_paired_degradation_statistics.csv`	48	28

Supporting files: source_bundle_checksums.json.

Corpus scope

Run	Model	Baseline	Quantized	Substrate	Licence
`Mistral_Nemo_Instruct_2407_20260814_030505`	mistralai/Mistral-Nemo-Instruct-2407	gguf_f16	gguf_q4_k_m	local	Apache-2.0
`Mistral_Nemo_Instruct_2407_20260815_113254`	mistralai/Mistral-Nemo-Instruct-2407	gguf_f16	gguf_q5_k_m	local	Apache-2.0
`Mistral_Nemo_Instruct_2407_20260816_084553`	mistralai/Mistral-Nemo-Instruct-2407	gguf_f16	gguf_q8_0	local	Apache-2.0
`Qwen2.5_14B_Instruct_1M_20260815_220633`	Qwen/Qwen2.5-14B-Instruct-1M	modal_f16	modal_q4_k_m	Modal	Apache-2.0
`Qwen2.5_32B_Instruct_20260815_081051`	Qwen/Qwen2.5-32B-Instruct	modal_f16	modal_q4_k_m	Modal	Apache-2.0
`Qwen2.5_7B_Instruct_20260814_234822`	Qwen/Qwen2.5-7B-Instruct	gguf_f16	gguf_q4_k_m	local	Apache-2.0

Every run evaluates a full-weight baseline and a quantized variant of the same model against the identical locked fixture set, case for case. Statistical comparison is paired: the two-sided exact McNemar test on per-case outcomes, with Wilson intervals on the rates.

Figures

Paired pass-rate difference (quantized minus full weight) with 95% confidence intervals, Mistral-Nemo-Instruct-2407, n = 200 paired cases per family. Solid markers indicate two-sided exact McNemar p < 0.05. Families significant: Q4_K_M 6 of 8; Q5_K_M 0 of 8; Q8_0 0 of 8. The shape is a discontinuity, not a gradient.

Paired pass-rate difference at Q4_K_M versus each model's own F16 baseline, n = 200 paired cases per family. Solid markers indicate McNemar p < 0.05. CONFOUNDED: 7B ran on local llama.cpp with a fixed seed; 14B-1M and 32B ran on Modal, which records seed status unsupported. Scale and substrate are not separated by this corpus, and this figure must not be read as a clean parameter-scale effect.

Figures are generated directly from the harness rollups by the published build tooling; no plotted value is recomputed, smoothed, or fitted.

Columns

`quant_eval_paired_degradation_statistics.csv`

  run_id                                        model_id                                      baseline_quant_type
  quantized_quant_type                          family                                        runner_a
  runner_b                                      paired_n                                      both_pass
  both_fail                                     runner_a_only_pass                            runner_b_only_pass
  disagreements                                 disagreement_rate                             pass_rate_delta_runner_b_minus_runner_a
  delta_ci_low                                  delta_ci_high                                 delta_ci_level
  delta_ci_method                               mcnemar_p_value                               mcnemar_discordant_pairs
  mcnemar_alternative                           discriminating                                cluster_adjusted_method
  cluster_adjusted_effective_paired_n           cluster_adjusted_delta                        cluster_adjusted_delta_ci_low
  cluster_adjusted_delta_ci_high

Verification

This corpus is derived from sanitized publication bundles produced by the quant_eval harness. It is designed to be checked rather than trusted:

source_bundle_checksums.json, included here, republishes, verbatim, the SHA-256 digest and byte length of every file in every source bundle. No source file was modified.
Before this file was written, the builder verified all 72 source-file digests and independently recomputed all 96 family x runner pass rates from the raw per-case rows, matching the harness rollups exactly.
Every row in this file is recomputable from the per-case results dataset (D1) using the gate definitions in data_dictionary.json, published with the per-case results dataset (D1) and the throughput telemetry dataset (D2). The pass counts, rates, and full discordance breakdown follow directly from the per-case outcomes; no intermediate artifact is required.

Limits you should know before using this

Decoding conditions are not uniform across models. Temperature follows each publisher's own model card, so cross-model comparison of absolute pass rates is confounded. Within-run pairing is unaffected, which is what the paired test requires. The conditions are published per row and per run so they can be filtered on.
Runs on the Modal substrate record seed status unsupported — the deployed method signature accepts no seed — so those runs are not exactly reproducible. Local runs applied a fixed seed.
Runtime figures are observed harness wall time on the recorded hardware and backends. They are not a controlled throughput benchmark and not a general claim about quantization performance at any precision on any hardware. Direction is published as an explicit label because not every measured pair is a speedup.
The fuzz family is an adaptive trajectory evaluated from identical starting fixtures. Its paired test compares complete case outcomes, not identical post-divergence prompts.
Calibration runs are not published. Runs that informed a published run are disclosed by identifier in calibration_lineage.csv, published in the run provenance dataset (D4), so the record is complete without releasing provisional numbers.

Citation

@dataset{pbh_quant_eval_d5,
  author    = {Hill, Patrick},
  title     = {quant_eval Paired degradation statistics},
  publisher = {PBH Applied Systems, LLC},
  year      = {2026},
  doi       = {10.5281/zenodo.22010557},
  note      = {Version DOI: 10.5281/zenodo.22010558},
  license   = {CC-BY-4.0}
}

Licence

Creative Commons Attribution 4.0 International (CC BY 4.0). See LICENSE. Commercial use is permitted; attribution is required.

This corpus describes third-party models and redistributes no model weights. Each evaluated model remains under its own licence, recorded per run in the run provenance dataset.

Produced by build_datasets.py 2.4.0 from quant_eval publication bundles. Built 2026-08-19.

Downloads last month: 38

Number of rows:

Total file size:

399 kB