Refactor turbomind (low-level abstractions)

[lzhangzz]

No description provided.

[lvhan028]

Evaluation results on some chat models

dataset	version	metric	mode	qwen2.5-0.5b-instruct-turbomind-wo	qwen2.5-7b-instruct-turbomind-wo	qwen2.5-32b-instruct-turbomind-wo	qwen2.5-72b-instruct-turbomind-wo	internlm2_5-7b-chat-turbomind-wo	llama-3_1-8b-instruct-turbomind-wo
Language	-	-	-	-	-	-	-	-	-
race-high	bd3f33	accuracy	gen	41.91	85.19	91.08	90.74	86.62	82.22
ARC-c	926652	accuracy	gen	42.71	90.51	95.25	96.61	88.81	87.46
BoolQ	1d56df	accuracy	gen	59.02	86.48	87.95	89.27	88.26	85.99
triviaqa_wiki_1shot	bc5f21	score	gen	30.81	53.27	61.35	72.54	65.16	79.43
nq_open_1shot	2e45e5	score	gen	8.50	17.48	20.89	27.20	22.63	34.90
mmmlu_lite	-	naive_average	gen	27.98	54.89	60.59	68.99	44.49	43.77
	-	-	-	-	-	-	-	-	-
General Reasoning	-	-	-	-	-	-	-	-	-
drop	3857b0	accuracy	gen	34.38	80.25	88.08	87.55	77.29	81.45
bbh	-	naive_average	gen	30.23	68.16	84.49	84.30	73.45	67.93
GPQA_diamond	5aeece	accuracy	gen	12.63	37.37	45.45	50.00	31.31	25.76
hellaswag	e42710	accuracy	gen	39.63	85.40	92.21	92.71	94.82	76.98
TheoremQA	-	-	-	-	-	-	-	-	-
musr_average	-	naive_average	gen	34.86	42.12	53.23	48.80	49.59	54.32
korbench_single	-	naive_average	gen	11.20	41.44	56.24	52.88	32.72	44.96
ARC_Prize_Public_Evaluation	872059	accuracy	gen	0.00	0.00	0.06	0.09	0.01	0.01
	-	-	-	-	-	-	-	-	-
Math Calculation	-	-	-	-	-	-	-	-	-
gsm8k	6e39a4	accuracy	gen	47.92	92.65	95.38	95.30	87.11	84.46
GaokaoBench	-	weighted_average	gen	32.34	80.69	90.41	90.46	78.25	49.58
math	11c4b5	accuracy	gen	32.86	73.60	80.96	81.64	61.04	49.52
cmo_fib	ace24b	accuracy	gen	1.92	24.52	31.25	33.65	12.98	5.77
aime2024	6e39a4	accuracy	gen	0.00	13.33	23.33	26.67	3.33	3.33
Mathbench	-	naive_average	gen	19.25	77.42	84.62	85.03	64.10	51.92
	-	-	-	-	-	-	-	-	-
Knowledge	-	-	-	-	-	-	-	-	-
wikibench-wiki-single_choice_cncircular	0978ad	perf_4	gen	7.60	34.95	44.70	49.85	31.45	26.90
cmmlu	-	naive_average	gen	31.59	75.83	82.91	84.85	74.51	53.69
mmlu	-	naive_average	gen	34.86	76.31	84.24	86.38	70.69	71.71
mmlu_pro	-	naive_average	gen	13.65	56.19	68.20	71.40	45.27	48.30
	-	-	-	-	-	-	-	-	-
mmlu	-	naive_average	gen	34.86	76.31	84.24	86.38	70.69	71.71
mmlu-stem	-	naive_average	gen	30.27	78.26	86.04	87.83	68.84	67.58
mmlu-social-science	-	naive_average	gen	39.91	79.03	87.15	88.63	75.85	77.15
mmlu-humanities	-	naive_average	gen	34.66	72.13	82.67	85.51	68.33	71.81
mmlu-other	-	naive_average	gen	37.09	75.12	80.48	83.04	70.97	72.62
	-	-	-	-	-	-	-	-	-
cmmlu	-	naive_average	gen	31.59	75.83	82.91	84.85	74.51	53.69
cmmlu-stem	-	naive_average	gen	24.78	72.84	80.49	82.44	66.46	47.99
cmmlu-social-science	-	naive_average	gen	31.93	75.58	83.19	84.54	75.59	53.51
cmmlu-humanities	-	naive_average	gen	34.72	76.89	84.10	86.62	78.95	55.63
cmmlu-other	-	naive_average	gen	36.11	78.68	84.21	86.51	78.20	58.72
cmmlu-china-specific	-	naive_average	gen	32.38	72.70	80.79	82.93	73.48	48.02
	-	-	-	-	-	-	-	-	-
mmlu_pro	-	naive_average	gen	13.65	56.19	68.20	71.40	45.27	48.30
mmlu_pro_biology	58fe7c	accuracy	gen	26.08	73.36	84.10	82.71	68.62	68.76
mmlu_pro_business	58fe7c	accuracy	gen	10.52	68.06	75.03	78.83	49.43	52.34
mmlu_pro_chemistry	58fe7c	accuracy	gen	7.16	57.86	70.32	73.85	35.78	38.69
mmlu_pro_computer_science	58fe7c	accuracy	gen	12.68	60.00	72.20	76.59	48.29	48.54
mmlu_pro_economics	58fe7c	accuracy	gen	22.51	64.81	76.90	78.91	56.64	58.29
mmlu_pro_engineering	58fe7c	accuracy	gen	6.09	42.62	56.14	57.17	30.55	26.42
mmlu_pro_health	58fe7c	accuracy	gen	13.94	54.77	66.26	71.03	46.58	57.58
mmlu_pro_history	58fe7c	accuracy	gen	10.76	46.98	58.79	64.57	41.47	46.98
mmlu_pro_law	58fe7c	accuracy	gen	11.53	27.88	41.60	47.77	23.89	32.52
mmlu_pro_math	58fe7c	accuracy	gen	10.51	73.21	81.87	83.12	54.18	52.04
mmlu_pro_philosophy	58fe7c	accuracy	gen	13.83	44.29	59.12	62.12	38.48	41.28
mmlu_pro_physics	58fe7c	accuracy	gen	8.93	57.66	72.44	74.52	37.11	42.34
mmlu_pro_psychology	58fe7c	accuracy	gen	22.81	62.28	75.06	77.32	57.64	61.78
mmlu_pro_other	58fe7c	accuracy	gen	13.74	52.81	64.94	71.10	45.13	48.59
	-	-	-	-	-	-	-	-	-
mmmlu_lite	-	naive_average	gen	27.98	54.89	60.59	68.99	44.49	43.77
openai_mmmlu_lite_AR-XY	07891e	accuracy	gen	22.67	38.81	4.70	47.51	17.05	19.09
openai_mmmlu_lite_BN-BD	0c33f9	accuracy	gen	9.12	42.25	12.77	68.63	27.23	29.05
openai_mmmlu_lite_DE-DE	2f4124	accuracy	gen	31.58	60.21	74.95	75.72	50.88	26.46
openai_mmmlu_lite_ES-LA	555bbc	accuracy	gen	34.25	66.53	76.21	76.91	56.77	56.84
openai_mmmlu_lite_FR-FR	97f4e3	accuracy	gen	34.39	67.09	78.25	79.37	58.18	56.70
openai_mmmlu_lite_HI-IN	94cd28	accuracy	gen	19.09	50.25	65.05	73.75	30.11	51.30
openai_mmmlu_lite_ID-ID	b65293	accuracy	gen	28.42	61.68	75.44	77.26	50.81	53.54
openai_mmmlu_lite_IT-IT	bf8b68	accuracy	gen	33.19	65.33	77.05	79.44	50.74	56.63
openai_mmmlu_lite_JA-JP	45635b	accuracy	gen	29.68	61.12	73.96	77.75	50.39	53.40
openai_mmmlu_lite_KO-KR	bd4d2a	accuracy	gen	29.89	61.19	72.70	74.32	43.86	51.65
openai_mmmlu_lite_PT-BR	0b103d	accuracy	gen	28.84	54.88	71.16	66.60	57.40	24.28
openai_mmmlu_lite_SW-KE	75318c	accuracy	gen	26.95	36.49	49.40	49.40	32.49	43.30
openai_mmmlu_lite_YO-NG	75318c	accuracy	gen	25.82	32.49	37.75	39.44	31.79	31.86
openai_mmmlu_lite_ZH-CN	14e7b8	accuracy	gen	37.82	70.11	78.81	79.72	65.12	58.67
	-	-	-	-	-	-	-	-	-
###### MathBench-A: Application Part ######	-	-	-	-	-	-	-	-	-
college	-	naive_average	gen	4.00	47.33	66.33	65.00	19.33	16.00
high	-	naive_average	gen	13.33	59.67	70.67	70.00	47.33	28.33
middle	-	naive_average	gen	18.67	78.33	84.67	87.33	59.67	43.67
primary	-	naive_average	gen	20.00	84.33	87.33	88.33	73.33	65.00
arithmetic	-	naive_average	gen	36.00	75.00	79.33	82.33	58.00	59.00
mathbench-a (average)	-	naive_average	gen	18.40	68.93	77.67	78.60	51.53	42.40
###### MathBench-T: Theory Part ######	-	-	-	-	-	-	-	-	-
college_knowledge	-	naive_average	gen	13.29	83.70	90.35	90.66	68.04	62.34
high_knowledge	-	naive_average	gen	14.04	80.04	87.78	88.46	70.87	54.12
middle_knowledge	-	naive_average	gen	22.41	85.55	92.86	91.22	79.26	59.42
primary_knowledge	-	naive_average	gen	30.70	94.34	95.28	95.52	88.49	69.89
mathbench-t (average)	-	naive_average	gen	20.11	85.91	91.57	91.47	76.66	61.44