Datasetpublic

andyx10/ep-dashboards-qwen27b-p-deprecated

EP Dashboards — Qwen3.6-27B Exemplar-Partitioning dashboards for Qwen/Qwen3.6-27B over the Pile Contents layer p θ regions (K) dictionary 15 8 0.7977 13,627 qwen3_6-27b_L15_p8p0_ctx128_cache_pile 23 2 0.7932 20,887 qwen3_6-27b_L23_p2p0_ctx128_cache_pile 3 0.8316 8,691 qwen3_6-27b_L23_p3p0_ctx128_cache_pile 4 0.8547 4,761 qwen3_6-27b_L23_p4p0_ctx128_cache_pile 8 0.9003 1,096 qwen3_6-27b_L23_p8p0_ctx128_cache_pile 31 2 0.8170 13,597… See the full description on the dataset page: https://huggingface.co/datasets/andyx10/ep-dashboards-qwen27b-p-deprecated.

sourceHugging Facemitupdated 2mo agoView on Hugging Face

0likes144downloads

Dataset Card

EP Dashboards — Qwen3.6-27B

Exemplar-Partitioning dashboards for Qwen/Qwen3.6-27B over the Pile

layer	p	θ	regions (K)	dictionary
15	8	0.7977	13,627	`qwen3_6-27b_L15_p8p0_ctx128_cache_pile`
23	2	0.7932	20,887	`qwen3_6-27b_L23_p2p0_ctx128_cache_pile`
	3	0.8316	8,691	`qwen3_6-27b_L23_p3p0_ctx128_cache_pile`
	4	0.8547	4,761	`qwen3_6-27b_L23_p4p0_ctx128_cache_pile`
	8	0.9003	1,096	`qwen3_6-27b_L23_p8p0_ctx128_cache_pile`
31	2	0.8170	13,597	`qwen3_6-27b_L31_p2p0_ctx128_cache_pile`
	3	0.8497	4,499	`qwen3_6-27b_L31_p3p0_ctx128_cache_pile`
	4	0.8699	2,093	`qwen3_6-27b_L31_p4p0_ctx128_cache_pile`
	8	0.9110	400	`qwen3_6-27b_L31_p8p0_ctx128_cache_pile`
39	2	0.7601	33,111	`qwen3_6-27b_L39_p2p0_ctx128_cache_pile`
	3	0.8069	11,123	`qwen3_6-27b_L39_p3p0_ctx128_cache_pile`
	4	0.8371	4,919	`qwen3_6-27b_L39_p4p0_ctx128_cache_pile`
	8	0.8948	742	`qwen3_6-27b_L39_p8p0_ctx128_cache_pile`
47	2	0.7832	11,865	`qwen3_6-27b_L47_p2p0_ctx128_cache_pile`
	3	0.8181	4,558	`qwen3_6-27b_L47_p3p0_ctx128_cache_pile`
	4	0.8410	2,318	`qwen3_6-27b_L47_p4p0_ctx128_cache_pile`
	8	0.8902	465	`qwen3_6-27b_L47_p8p0_ctx128_cache_pile`
55	2	0.7400	30,258	`qwen3_6-27b_L55_p2p0_ctx128_cache_pile`
	3	0.7887	11,329	`qwen3_6-27b_L55_p3p0_ctx128_cache_pile`
	4	0.8218	5,190	`qwen3_6-27b_L55_p4p0_ctx128_cache_pile`
	8	0.8870	779	`qwen3_6-27b_L55_p8p0_ctx128_cache_pile`
	12	0.9163	280	`qwen3_6-27b_L55_p12p0_ctx128_cache_pile`
	16	0.9347	144	`qwen3_6-27b_L55_p16p0_ctx128_cache_pile`
	—	0.7726	16,051	`qwen3_6-27b_L55_k16051_ctx128_cache_pile`
56	2	0.7243	40,591	`qwen3_6-27b_L56_p2p0_ctx128_cache_pile`
	3	0.7763	15,347	`qwen3_6-27b_L56_p3p0_ctx128_cache_pile`
	4	0.8120	7,008	`qwen3_6-27b_L56_p4p0_ctx128_cache_pile`
	8	0.8811	1,014	`qwen3_6-27b_L56_p8p0_ctx128_cache_pile`
62	2	0.7284	44,725	`qwen3_6-27b_L62_p2p0_ctx128_cache_pile`
	3	0.7786	17,019	`qwen3_6-27b_L62_p3p0_ctx128_cache_pile`
	4	0.8124	7,684	`qwen3_6-27b_L62_p4p0_ctx128_cache_pile`
	8	0.8791	1,110	`qwen3_6-27b_L62_p8p0_ctx128_cache_pile`

*k16051, test dict for fixing region count near 16k

Each dictionary folder holds:

file	contents
`regions_NNN.html`	region pages — EPDashboard's unit of output, JSON embedded
`regions_NNN.json`	the region records those pages embed
`header.json`	dict metadata, provenance, replay check, per-region summary table
`vectors.npz`	`exemplar` + `mean` direction per region, plus the calibration `center`

Generated by EPDashboard.