Datasets:

zelk12
/

BenchLocal_Result

Model name stringlengths 15 42	Model URL stringlengths 38 65	Model Quantization stringclasses 4 values	Average float64 9.92 86	InstructFollow-15 float64 5.8 87.6	Format Constraints int64 0 100	Ordering and Sorting int64 7 100	Multi-Domain int64 7 100	Precision Under Pressure int64 15 80	Adversarial int64 0 100	DataExtract-15 float64 0 86.8	Clean Extraction int64 0 97	Noisy and Informal int64 0 86	Multi-Entity int64 0 92	Implicit and Missing int64 0 91	Complex Documents int64 0 88	ToolCall-15 float64 0 96.6	Tool Selection int64 0 100	Parameter Precision int64 0 100	Multi-Step Chains int64 0 100	Restraint & Refusal int64 0 100	Error Recovery int64 0 100	ReasonMath-15 float64 0 80.8	Everyday Arithmetic int64 0 100	Logic Puzzles int64 0 62	Multi-Step Word Problems int64 0 85	Trick Questions and Traps int64 0 100	Applied Reasoning int64 0 85	PromptAuthority-15 float64 6.6 93.4	Hard System vs User int64 0 100	User Overrides Defaults int64 33 100	User vs Tool Data int64 0 100	Hard System vs Tool Data int64 0 100	Layered Conflicts int64 0 100
google/gemma-4-E2B-it	https://huggingface.co/google/gemma-4-E2B-it	Q8	68.44	82.4	72	100	100	73	67	74.8	89	65	85	82	53	73.4	83	67	67	83	67	71.6	77	62	85	72	62	40	33	100	67	0	0
google/gemma-4-E4B-it	https://huggingface.co/google/gemma-4-E4B-it	Q6	80.36	85.6	83	100	100	78	67	84.2	85	86	92	80	78	66.4	83	100	33	83	33	78.8	100	52	85	95	62	86.8	100	100	100	67	67
google/gemma-4-12B-it	https://huggingface.co/google/gemma-4-12B-it	Q4	63.24	39.2	0	47	67	15	67	48.4	97	0	63	49	33	90	100	67	100	83	100	58.4	77	20	85	72	38	80.2	100	67	100	67	67
google/gemma-4-26B-A4B-it	https://huggingface.co/google/gemma-4-26B-A4B-it	Q2	56.16	45.6	0	80	33	15	100	30	54	0	63	0	33	90	100	100	67	83	100	48	77	10	62	53	38	67.2	100	100	33	3	100
google/gemma-4-31B-it	https://huggingface.co/google/gemma-4-31B-it	Q2	85.96	83.8	100	100	67	52	100	83.4	92	62	92	91	80	93.2	100	100	100	83	83	76	77	38	85	95	85	93.4	100	100	100	67	100
Qwen/Qwen3.5-0.8B	https://huggingface.co/Qwen/Qwen3.5-0.8B	Q8	21.04	5.8	0	7	7	15	0	0	0	0	0	0	0	63.6	67	67	67	67	50	16	30	10	15	20	5	19.8	0	33	0	33	33
Qwen/Qwen3.5-2B	https://huggingface.co/Qwen/Qwen3.5-2B	Q8	39.76	24.2	33	40	33	15	0	6.6	0	0	0	0	33	96.6	100	100	100	83	100	44.8	48	33	62	48	33	26.6	33	67	33	0	0
Qwen/Qwen3.5-9B	https://huggingface.co/Qwen/Qwen3.5-9B	Q4	70.16	69.8	67	100	67	15	100	46.8	64	24	84	29	33	93.4	100	100	100	100	67	80.8	100	62	85	95	62	60	67	67	100	33	33
Nanbeige/Nanbeige4.1-3B	https://huggingface.co/Nanbeige/Nanbeige4.1-3B	Q8	64.64	58.6	80	47	67	32	67	35.2	31	0	57	55	33	86.8	67	67	100	100	100	76	100	38	85	95	62	66.6	100	67	33	100	33
Nanbeige/Nanbeige4.2-3B	https://huggingface.co/Nanbeige/Nanbeige4.2-3B	Q8	76.56	87.6	80	80	100	78	100	70.4	82	65	60	91	54	93.4	100	100	100	67	100	71.4	77	33	85	100	62	60	67	100	67	33	33
allura-forge/Llama-3.3-8B-Instruct	https://huggingface.co/allura-forge/Llama-3.3-8B-Instruct	Q6	39.68	55.6	63	48	74	53	40	63	58	65	57	68	67	16.8	17	0	0	67	0	36.4	15	5	62	67	33	26.6	33	100	0	0	0
EssentialAI/rnj-1-instruct	https://huggingface.co/EssentialAI/rnj-1-instruct	Q6	46.04	68.8	47	93	100	43	61	71.6	80	51	84	71	72	13.4	0	0	0	67	0	56.4	77	33	62	72	38	20	0	100	0	0	0
mistralai/Ministral-3-3B-Instruct-2512	https://huggingface.co/mistralai/Ministral-3-3B-Instruct-2512	Q8	37.44	58.4	75	45	88	65	19	0	0	0	0	0	0	56.6	83	33	67	83	17	52.2	48	28	85	67	33	20	0	100	0	0	0
mistralai/Ministral-3-8B-Instruct-2512	https://huggingface.co/mistralai/Ministral-3-8B-Instruct-2512	Q6	40.32	55.4	78	22	88	53	36	0	0	0	0	0	0	33.2	17	0	33	83	33	73.2	100	52	85	67	62	39.8	33	100	0	33	33
mistralai/Ministral-3-3B-Reasoning-2512	https://huggingface.co/mistralai/Ministral-3-3B-Reasoning-2512	Q8	42.48	66.2	68	32	77	67	87	0	0	0	0	0	0	63.4	67	33	67	83	67	49.6	43	5	85	53	62	33.2	33	100	33	0	0
mistralai/Ministral-3-8B-Reasoning-2512	https://huggingface.co/mistralai/Ministral-3-8B-Reasoning-2512	Q6	45.8	55.4	63	32	93	53	36	0	0	0	0	0	0	70	83	67	33	100	67	70.4	100	33	85	72	62	33.2	33	100	0	0	33
google/gemma-3-270m-it	https://huggingface.co/google/gemma-3-270m-it	Q8	11.72	31.8	45	22	38	37	17	0	0	0	0	0	0	13.4	0	0	0	67	0	0	0	0	0	0	0	13.4	0	67	0	0	0
tiiuae/Falcon-H1-0.5B-Instruct	https://huggingface.co/tiiuae/Falcon-H1-0.5B-Instruct	Q8	9.92	26.4	32	15	38	28	19	0	0	0	0	0	0	0	0	0	0	0	0	16.6	15	5	5	48	10	6.6	0	33	0	0	0
tiiuae/Falcon-H1-1.5B-Instruct	https://huggingface.co/tiiuae/Falcon-H1-1.5B-Instruct	Q8	20.08	30.8	40	38	31	28	17	0	0	0	0	0	0	6.6	0	0	0	33	0	49.6	48	28	62	72	38	13.4	0	67	0	0	0
tiiuae/Falcon-H1-1.5B-Deep-Instruct	https://huggingface.co/tiiuae/Falcon-H1-1.5B-Deep-Instruct	Q8	26.2	30.8	40	38	43	22	11	0	0	0	0	0	0	20	0	0	0	100	0	53.4	100	5	62	90	10	26.8	67	67	0	0	0
tiiuae/Falcon-H1-3B-Instruct	https://huggingface.co/tiiuae/Falcon-H1-3B-Instruct	Q8	24.16	33.2	40	38	43	28	17	0	0	0	0	0	0	13.4	0	0	0	67	0	61	100	33	62	77	33	13.2	33	33	0	0	0
tiiuae/Falcon-H1-7B-Instruct	https://huggingface.co/tiiuae/Falcon-H1-7B-Instruct	Q6	22.72	36.4	40	45	38	48	11	0	0	0	0	0	0	6.6	0	0	0	33	0	43.8	20	28	85	53	33	26.8	67	67	0	0	0
upstage/SOLAR-10.7B-Instruct-v1.0	https://huggingface.co/upstage/SOLAR-10.7B-Instruct-v1.0	Q4	14.96	27	47	7	23	33	25	20.2	14	28	23	3	33	13.4	0	0	0	67	0	7.6	0	28	0	5	5	6.6	0	33	0	0	0
nvidia/NVIDIA-Nemotron-Nano-9B-v2	https://huggingface.co/nvidia/NVIDIA-Nemotron-Nano-9B-v2	Q4	70.28	79.8	72	100	93	67	67	83	92	72	89	85	77	76.8	100	67	67	83	67	58.6	67	23	85	90	28	53.2	33	100	33	100	0
nvidia/NVIDIA-Nemotron-Nano-12B-v2-VL-BF16	https://huggingface.co/nvidia/NVIDIA-Nemotron-Nano-12B-v2-VL-BF16	Q4	50.88	85.4	72	92	94	80	89	15.6	0	26	52	0	0	66.6	83	33	67	83	67	47	53	10	38	72	62	39.8	33	100	33	33	0
stamsam/MedusaGemma-E4B	https://huggingface.co/stamsam/MedusaGemma-E4B	Q6	71.16	78.4	80	72	93	53	94	86.8	95	86	87	78	88	60.2	67	67	33	67	67	70.4	95	38	85	72	62	60	33	100	67	67	33

Описание/Description

EN

Here are the results of model testing using the BenchLocal program, the following settings were used: min_p = 1 request time out = 10 000. Five categories were launched for the test, each containing 15 tasks. This verification method checks two things at once: first, that the language model is capable of performing the assigned task. Second, that the model doesn't get stuck or freeze during reasoning.

RU

Здесь представлены результаты проверки моделей, через программу BenchLocal, были использованны следующие настройки, min_p = 1 request time out = 10 000. Для теста запущенно 5 категорий, в каждой 15 задач. Данный вариант проверки проверяет сразу 2 вещи, первое, это то, что языковая модель, способна выполнить поставленную задачу. Второе, это что модель не зацикливаеться и не зависает во время мышления.

Downloads last month: 118

Number of rows:

Total file size:

9.14 kB