Leaderboard
Subnet 9 - AI Pretraining

Subnet 9 is a Bittensor subnet that rewards miners for producing pretrained Foundation-Models on the Hugging Face's FineWeb Edu (score 2) dataset.

It acts like a continuous benchmark whereby miners are rewarded for attaining the best losses on randomly sampled pages.

The models with the best head-to-head win-rate on the evaluation data receive a steady emission of TAO.

Links:

Discord Github Whitepaper

Competitions

This is a Multi-Dataset competition.

Model improvement over time

Live Leaderboard

2nd

deepnet111 / sn9-...

UID 5

Incentive0.0000Emission0.0000Trust0.0000Average Loss1.9698

1st

tensoralchemistde...

UID 148

Incentive0.0000Emission0.0000Trust0.0000Average Loss1.9654

3rd

rwh / multi23

UID 196

Incentive0.0000Emission0.0000Trust0.0000Average Loss2.1047

Rank
1st	tensoralchemistdev01 / bb18	0.521	0.209	1.965	0.3564	0.67	LlamaForCausalLM	f9a39c4654f13df81f96fa4c876a15a0d50a2104	148	4786562
2nd	deepnet111 / sn9-14b-star-016	0.239	0.096	1.97	0.3365	0.67	LlamaForCausalLM	d5fcd248d84f16ea62e7a9a151c9a090316da8d0	5	4900987
3rd	rwh / multi23	0	0	2.105	0.02	0.75	-	77e72d8651d3b629697acc3d6655a9972734b14a	196	4849560
4	rwh / multi13	0	0	2.017	0.02	0	-	c191cbe3bddf0b4bad66f89fca58487b1f001e9c	185	4792412
5	rwh / coder04	0	0	2.079	0.02	0.33	SlicedLlamaForCausalLM	cb024a2ee8e9d71b914202c37c0735e80cc9649b	203	4450296
6	jw-hf-test / jw-14bstar-002	0	0	1.849	0.3896	0.75	LlamaForCausalLM	663adfbf9b72519a1909366a9e585a503247d123	27	4486347
7	jw-hf-test / jw-14B-211	0	0	2.072	0.02	0	LlamaForCausalLM	21a676132f4bdf1a6d0970569d30c87a6bac5fdd	30	4039523
8	rwh / coder06	0	0	2.048	0.02	0.75	SlicedLlamaForCausalLM	e55937f13d463e3daf1770ad306086bed300bb97	206	4464734
9	rwh / multi15	0	0	1.996	0.02	0.75	-	fb33d484704745f989d806c409e01c10955d37ec	190	4797650
10	rwh / coder03	0	0	2.056	0.02	0.33	SlicedLlamaForCausalLM	bf4e68a954cb25fccdb66609f2b4395b047e172f	199	4450033
11	visdata / ss0	0	0	1.964	0.41	0	LlamaForCausalLM	4a721aae80ef4aa9e74c6b490699b5e6cc8355dc	166	4919439
12	rwh / coder09	0	0	1.917	0.02	0.67	SlicedLlamaForCausalLM	907da752f308f7b1945c30e6063d886e27837382	193	4492260
13	jw-hf-test / jw-14B-300	0	0	2.16	0.02	0	LlamaForCausalLM	fa70fa9acabae2f7561bed154125d9b3aed3aeda	8	4646444
14	jw-hf-test / jw-14B-220	0	0	2.109	0.02	0.5	LlamaForCausalLM	81f19366fd0e164298ebf83c708038e153fc0556	25	4485603
15	jw-hf-test / jw-14B-219	0	0	2.103	0.02	0	LlamaForCausalLM	a8c5439c4c4fdd79127c019111df9e77f71ca8c6	69	4392751
16	axon2 / sn9_14b_106	0	0	1.894	0.3446	0	LlamaForCausalLM	64be0e11243215086972b50faba09c01cd99d616	106	4769589
17	rwh / h005	0	0	2.071	0.02	0	SlicedLlamaForCausalLM	b2bc15651e0bca57e4f8a2b0e57b289603b11469	202	4511390
18	jw-hf-test / jw-14B-218	0	0	2.021	0.02	0	LlamaForCausalLM	dd28a9d53955ea6aa794faa523c218c4323ee152	66	4291087
19	jw-hf-test / jw-14B-215	0	0	2.111	0.02	0	LlamaForCausalLM	62c495f181db4880336c15e29175304a52c06597	2	4199668
20	rwh / multi21	0	0	2.131	0.05	0	-	3871388e08304969e4dc5ebb2f447dc516178083	194	4842455
21	jw-hf-test / jw-14bstar-001	0	0	1.931	0.02	0	LlamaForCausalLM	0bcbe76da9691aa6d4409f9a4ba5fd2ad81e6d7b	114	4394129
22	jw-hf-test / jw-14B-214	0	0	2.11	0.02	0	LlamaForCausalLM	7f0ee237035437bc779d05e1672f87602f9f9303	16	4149650
23	jw-hf-test / jw-14B-216	0	0	2.131	0.02	0	LlamaForCausalLM	50c4ddc98c241e46c06bb594df413c0b74110b5e	252	4222047
24	visdata / bb_01	0	0	2.006	0.02	0	LlamaForCausalLM	c129dff0e0007f18effcc944746540cbeeb34228	82	4658280
25	jw-hf-test / jw-14B-302	0	0	2.399	0.02	0	LlamaForCausalLM	1627663b233996b61de432940d3d3dd147662d48	80	4688895
26	rwh / multi16	0	0	2.082	0.05	0.25	-	9ae5bdb52ff4ccd05083a1514fab8b4c6cd5c6df	188	4809595
27	deepnet / SN9-C4-llama-HK5-1	0	0	2.144	0.02	0.17	LlamaForCausalLM	fc51dd83c5634f8ddab2c6101e428674e5ae5397	127	4136683
28	rwh / multi06	0	0	2.154	0.05	0	SlicedLlamaForCausalLM	6fdb5e068666a204b326f288e4d05c156c1d2efb	183	4766439
29	rwh / h003	0	0	2.066	0.02	0	SlicedLlamaForCausalLM	129b59fa074d165618be6b8644711798b21dddbb	204	4450152
30	aipretrained / SN9_T1101FF06	0	0	2.079	0.02	0	LlamaForCausalLM	9753d7a5c018fbe7a9c600efccfa75e296817072	21	4057184
31	aipretrained / SN9_T11010004	0	0	2.118	0.05	0	LlamaForCausalLM	952bec4cb3ad192f379bfed51b72eb558182095c	58	4056834
32	visdata / sn9-CL11	0	0	2.104	0.4849	0	LlamaForCausalLM	d5520e5b6dad9215f2d6d43cdc59cf03b4e86cc2	107	4919456
33	deepnet111 / sn9-14b-star-015	0	0	2.003	0.3453	0	LlamaForCausalLM	a54060f803985b7cf2ca37894d21d81d029f2630	12	4879706
34	rwh / multi14	0	0	2.123	0.05	0	-	e676037e1c60deefa82174bc446f19a650e57870	181	4795297
35	rwh / multi09	0	0	2.054	0.05	0	-	8bf623f400f5bf1effe64446170281aa757f6097	182	4773510
36	yonderjay / house-11-14-04	0	0	2.137	0.02	0	LlamaForCausalLM	21df74899d6dcad9e372a237b7110677f8e5b5b2	54	4271335
37	jw-hf-test / jw-14B-400	0	0	2.113	0.4897	0	LlamaForCausalLM	2b97ef776cb047eddabc8eed43dca2f2335adbc6	144	4701534
38	visdata / ban16	0	0	2.086	0.4029	0	LlamaForCausalLM	bebb1362314361b1deae341818806469850bebaa	131	4915255
39	visdata / ban2	0	0	2.077	0.4504	0	LlamaForCausalLM	329bf0584501485da608ebaf3aa3d8154089b368	129	4915227
40	jw-hf-test / jw-14B-221	0	0	2.09	0.3126	0	LlamaForCausalLM	6f7c2dae1bf9adc72a7db41c840090f837be5de2	88	4517485
41	luaqi / sn9_fat_v2	0	0	2.14	0.02	0	LlamaForCausalLM	4b626fad56b5a14da0cebb518454107007d6dd6e	221	3754511
42	visdata / sn9-MG4	0	0	2.041	0.1616	0	LlamaForCausalLM	56a321bc69798541bcdd91fd9793c74371971890	96	4919466
43	jw-hf-test / jw-14B-303	0	0	2.076	0.3131	0	LlamaForCausalLM	c1dae6013a8d9b80db2fdbead69a58e0a2f96cb8	18	4695462
44	visdata / sn9-MG2	0	0	1.992	0.492	0	LlamaForCausalLM	63e3fa34a703e3452f277f747d73f22fa4f4626e	50	4919477
45	visdata / sn9-MG6	0	0	1.969	0.493	0	LlamaForCausalLM	d69c51353e173e12db2952b304527d701d0d88e1	130	4919471
46	visdata / sn9-MG3	0	0	1.962	0.49	0	LlamaForCausalLM	c46e847aa85cc0a5fd392961c9de5cf559f81e64	108	4919453
47	visdata / mmt3	0	0	5.051	0.4937	0	LlamaForCausalLM	015a1c398ef5090757de0e2bb3721b721a74cf08	40	4919441
48	visdata / mmt8	0	0	1.969	0.493	0	LlamaForCausalLM	352afbc9a28e310859396448b756a91580e86117	167	4919439
49	visdata / cnn10	0	0	2.758	0.4944	0	LlamaForCausalLM	c21ab57065e490fcdb69f5ba8c0e67e65d173700	159	4919437
50	visdata / sn9-CL0	0	0	2.004	0.4927	0	LlamaForCausalLM	42c949a34e501b0ba01be08670d82a5fbf9ca422	121	4919432

Live Per-dataset Leaderboard

Daily Perplexity Benchmarks

Daily SOTA Benchmarks

No benchmarks available for this competition.

Leaderboard Subnet 9 - AI Pretraining