Xunzhuo's picture
Expand encoder retrieval coverage with matching exports and measured scores
972c180 verified
Raw History Blame Contribute Delete
14.7 kB
{
"task": "LEMBSummScreenFDRetrieval",
"split": "validation",
"subset": "default",
"main_metric": "ndcg_at_10",
"scores": {
"original": 0.76988,
"vela": 0.8323,
"r6": 0.88493,
"r9": 0.87688,
"r13": 0.88041,
"r20": 0.88471,
"candidate": 0.88362,
"r3": 0.86925
},
"score_scale": "Official native FP32 metrics on0–1 scale; multiply by100 for display.",
"candidate_content_sha256": "3bf3ad85cdc82f8c5e67d44a59088b89de98f465a1a2b9126ff43a5f6f4b6225",
"dataset_revision": "6e346642246bfb4928c560ee08640dc84d074e8c",
"coverage": {
"queries": 336,
"documents": 336,
"configured_max_length": 32768,
"document_token_lengths": {
"minimum": 2662,
"maximum": 23856,
"sum": 2887533,
"count": 336
},
"truncated_documents": 0
},
"paired_comparison": {
"status": "completed",
"gate_component": {
"passed": true,
"delta_vs_r13": 0.0032154935251345294,
"minimum_delta": -0.005,
"units": "raw score delta",
"release_decision": false
},
"scope": "Already exposed long-development validation. No new model inference; no sealed confirmation data.",
"candidate_content_sha256": "3bf3ad85cdc82f8c5e67d44a59088b89de98f465a1a2b9126ff43a5f6f4b6225",
"task": "LEMBSummScreenFDRetrieval",
"split": "validation",
"metric": "ndcg_at_10",
"replicates": 2000,
"seed": 20260915,
"unit": "Paired query/source-document, verified336distinct sources",
"evidence": {
"original": {
"official_ndcg_at_10": 0.76988,
"recomputed_ndcg_at_10": 0.7698826358058495,
"result_sha256": "2e3c8594a8b7f7f62fad52f7dceaaf338b1146f8d90ad731b6345d50c6ffe833",
"prediction_sha256": "e14f41b52de7d7a9f6956283a9dee3dcdf496f42c90f824153922d04ca4ef743",
"qrels_sha256": "e8e521e2122fedc7d57678af6819ae09418cc7e346b6f029099da1a606c64558",
"token_lengths_sha256": "e8c0d697b5a37c08a9e3ed177295a6e6a23fd15e2e81adf0e17f1a4dc88f7d93"
},
"vela": {
"official_ndcg_at_10": 0.8323,
"recomputed_ndcg_at_10": 0.8323025262730762,
"result_sha256": "2a0b49ad538fa71dccd5bde5bdbd0aa3d877708f7ac6cc3ea435852c7812ddf8",
"prediction_sha256": "9340d48309f598ab2ee5ba3b58fcfe37a530f052528ead0f754ab37bea8fa030",
"qrels_sha256": "e8e521e2122fedc7d57678af6819ae09418cc7e346b6f029099da1a606c64558",
"token_lengths_sha256": "3db940cd46ac4b4d4c62af8d98baa1b116a761305f2005e2c658e9bffe728862"
},
"r6-intent-all22-lr1e5": {
"official_ndcg_at_10": 0.88493,
"recomputed_ndcg_at_10": 0.884925127232939,
"result_sha256": "b7ea70458c34d2fd0f6210d19f95440fb33cafba551f1d404978cc8ca6200241",
"prediction_sha256": "8cb20cd7bd7442d7c80fb29c5cf54525a111eb6e67c4b509ed90314146da79f7",
"qrels_sha256": "e8e521e2122fedc7d57678af6819ae09418cc7e346b6f029099da1a606c64558",
"token_lengths_sha256": "cdd98ac1f6110610422007186dc6b19823466446aa6cfca379b45ac4d6b89f0c"
},
"r9-guide-mask-lr3e6": {
"official_ndcg_at_10": 0.87688,
"recomputed_ndcg_at_10": 0.876882778899065,
"result_sha256": "95f44d11d6ff02fd672cc2c0fe7398ee148177ffd5291277d14d305e625499b0",
"prediction_sha256": "7c5f26d11fcb98cb6efd07bc1f56f590b463dbda83b017867783c6cd3a29fc07",
"qrels_sha256": "e8e521e2122fedc7d57678af6819ae09418cc7e346b6f029099da1a606c64558",
"token_lengths_sha256": "0c78366b6fddb877399974ba0e3671d5f49cf14a49dc38db52126504e9ce5151"
},
"r13-bge-kd-w01": {
"official_ndcg_at_10": 0.88041,
"recomputed_ndcg_at_10": 0.8804095047305854,
"result_sha256": "34afebd3836a1d25bf0e7798a6c47d979ff4537d389a0fa46494f81ba3b3572f",
"prediction_sha256": "9d314706274078458ea080b56c7c63a2b3e339ea20584d980bc919cfc43c2dbf",
"qrels_sha256": "e8e521e2122fedc7d57678af6819ae09418cc7e346b6f029099da1a606c64558",
"token_lengths_sha256": "fd547245c7fdc2b299dd8a4e1bd7631ccee3fb299911933301cacacd6e7ad7ef"
},
"r20-t2ranking-zh8192": {
"official_ndcg_at_10": 0.88471,
"recomputed_ndcg_at_10": 0.8847132026337041,
"result_sha256": "469de4a61c0c787dfe3bfc52805fa51b6d22e94ca712682f46a5868b4f85f088",
"prediction_sha256": "801ab1f7fb36034b361cdc0996cfff3473ad4704292f301e0294b933566d1400",
"qrels_sha256": "e8e521e2122fedc7d57678af6819ae09418cc7e346b6f029099da1a606c64558",
"token_lengths_sha256": "2df6252ed766d521014b14ee0d9d9355c6285e7dcf7a87177daa44a3acdbf9a3"
},
"r24-new-queries": {
"official_ndcg_at_10": 0.88362,
"recomputed_ndcg_at_10": 0.8836249982557198,
"result_sha256": "fe279244d0ae16d2c759e4320e28b0b3e0f82688386d5aa429103430277e27b3",
"prediction_sha256": "4415bf7da963741204a32979abd479066736bb47fe6dda101a6e6618abeb4b14",
"qrels_sha256": "e8e521e2122fedc7d57678af6819ae09418cc7e346b6f029099da1a606c64558",
"token_lengths_sha256": "c28375f904225f4bc5d62415e676f732970fd04ca540b45119abb0a0e4bd6800"
}
},
"rows": [
{
"candidate": "r24-new-queries",
"baseline": "original",
"group": "all",
"queries": 336,
"candidate_score": 0.8836249982557198,
"baseline_score": 0.7698826358058495,
"delta": 0.11374236244987043,
"paired_bootstrap_95_interval": [
0.08580597893036003,
0.1436500045695695
],
"improved": 94,
"equal": 225,
"regressed": 17
},
{
"candidate": "r24-new-queries",
"baseline": "original",
"group": "relevant_document_at_most_8192_tokens",
"queries": 170,
"candidate_score": 0.9207402405116178,
"baseline_score": 0.839537458569051,
"delta": 0.08120278194256693,
"paired_bootstrap_95_interval": [
0.04316328424717932,
0.12102952443348348
],
"improved": 35,
"equal": 126,
"regressed": 9
},
{
"candidate": "r24-new-queries",
"baseline": "original",
"group": "relevant_document_above_8192_tokens",
"queries": 166,
"candidate_score": 0.845615412812933,
"baseline_score": 0.6985493835784745,
"delta": 0.14706602923445836,
"paired_bootstrap_95_interval": [
0.10448118709621954,
0.19084205323592268
],
"improved": 59,
"equal": 99,
"regressed": 8
},
{
"candidate": "r24-new-queries",
"baseline": "vela",
"group": "all",
"queries": 336,
"candidate_score": 0.8836249982557198,
"baseline_score": 0.8323025262730762,
"delta": 0.05132247198264383,
"paired_bootstrap_95_interval": [
0.02595699843539391,
0.0778629398066223
],
"improved": 59,
"equal": 246,
"regressed": 31
},
{
"candidate": "r24-new-queries",
"baseline": "vela",
"group": "relevant_document_at_most_8192_tokens",
"queries": 170,
"candidate_score": 0.9207402405116178,
"baseline_score": 0.8468191092768992,
"delta": 0.07392113123471859,
"paired_bootstrap_95_interval": [
0.04501357914334632,
0.10670490435922182
],
"improved": 30,
"equal": 135,
"regressed": 5
},
{
"candidate": "r24-new-queries",
"baseline": "vela",
"group": "relevant_document_above_8192_tokens",
"queries": 166,
"candidate_score": 0.845615412812933,
"baseline_score": 0.817436146088438,
"delta": 0.02817926672449499,
"paired_bootstrap_95_interval": [
-0.014128021736724831,
0.07003966358355486
],
"improved": 29,
"equal": 111,
"regressed": 26
},
{
"candidate": "r24-new-queries",
"baseline": "r6-intent-all22-lr1e5",
"group": "all",
"queries": 336,
"candidate_score": 0.8836249982557198,
"baseline_score": 0.884925127232939,
"delta": -0.0013001289772192758,
"paired_bootstrap_95_interval": [
-0.008658072287171298,
0.005251621338530261
],
"improved": 18,
"equal": 306,
"regressed": 12
},
{
"candidate": "r24-new-queries",
"baseline": "r6-intent-all22-lr1e5",
"group": "relevant_document_at_most_8192_tokens",
"queries": 170,
"candidate_score": 0.9207402405116178,
"baseline_score": 0.9245553524790817,
"delta": -0.0038151119674638572,
"paired_bootstrap_95_interval": [
-0.012504089298200982,
0.002641760875011354
],
"improved": 4,
"equal": 162,
"regressed": 4
},
{
"candidate": "r24-new-queries",
"baseline": "r6-intent-all22-lr1e5",
"group": "relevant_document_above_8192_tokens",
"queries": 166,
"candidate_score": 0.845615412812933,
"baseline_score": 0.8443399568001426,
"delta": 0.0012754560127902356,
"paired_bootstrap_95_interval": [
-0.009602981539440099,
0.012731905072227398
],
"improved": 14,
"equal": 144,
"regressed": 8
},
{
"candidate": "r24-new-queries",
"baseline": "r9-guide-mask-lr3e6",
"group": "all",
"queries": 336,
"candidate_score": 0.8836249982557198,
"baseline_score": 0.876882778899065,
"delta": 0.0067422193566549615,
"paired_bootstrap_95_interval": [
0.0006376803680880756,
0.01385482075292914
],
"improved": 19,
"equal": 311,
"regressed": 6
},
{
"candidate": "r24-new-queries",
"baseline": "r9-guide-mask-lr3e6",
"group": "relevant_document_at_most_8192_tokens",
"queries": 170,
"candidate_score": 0.9207402405116178,
"baseline_score": 0.9142852803660881,
"delta": 0.006454960145529764,
"paired_bootstrap_95_interval": [
-0.0024266134694441643,
0.01719387898700963
],
"improved": 6,
"equal": 162,
"regressed": 2
},
{
"candidate": "r24-new-queries",
"baseline": "r9-guide-mask-lr3e6",
"group": "relevant_document_above_8192_tokens",
"queries": 166,
"candidate_score": 0.845615412812933,
"baseline_score": 0.838579012336451,
"delta": 0.00703640047648197,
"paired_bootstrap_95_interval": [
-0.0023528919316009025,
0.016606970299408556
],
"improved": 13,
"equal": 149,
"regressed": 4
},
{
"candidate": "r24-new-queries",
"baseline": "r13-bge-kd-w01",
"group": "all",
"queries": 336,
"candidate_score": 0.8836249982557198,
"baseline_score": 0.8804095047305854,
"delta": 0.0032154935251345294,
"paired_bootstrap_95_interval": [
-0.002654427117917938,
0.009620371358263215
],
"improved": 16,
"equal": 311,
"regressed": 9
},
{
"candidate": "r24-new-queries",
"baseline": "r13-bge-kd-w01",
"group": "relevant_document_at_most_8192_tokens",
"queries": 170,
"candidate_score": 0.9207402405116178,
"baseline_score": 0.9174067251903617,
"delta": 0.00333351532125612,
"paired_bootstrap_95_interval": [
-0.004205604809041235,
0.012017521119574765
],
"improved": 5,
"equal": 162,
"regressed": 3
},
{
"candidate": "r24-new-queries",
"baseline": "r13-bge-kd-w01",
"group": "relevant_document_above_8192_tokens",
"queries": 166,
"candidate_score": 0.845615412812933,
"baseline_score": 0.8425207849826216,
"delta": 0.003094627830311214,
"paired_bootstrap_95_interval": [
-0.006415175063604415,
0.012801972665767698
],
"improved": 11,
"equal": 149,
"regressed": 6
},
{
"candidate": "r24-new-queries",
"baseline": "r20-t2ranking-zh8192",
"group": "all",
"queries": 336,
"candidate_score": 0.8836249982557198,
"baseline_score": 0.8847132026337041,
"delta": -0.0010882043779841832,
"paired_bootstrap_95_interval": [
-0.005630972134979018,
0.003024722199180693
],
"improved": 2,
"equal": 326,
"regressed": 8
},
{
"candidate": "r24-new-queries",
"baseline": "r20-t2ranking-zh8192",
"group": "relevant_document_at_most_8192_tokens",
"queries": 170,
"candidate_score": 0.9207402405116178,
"baseline_score": 0.9214058122337676,
"delta": -0.0006655717221497551,
"paired_bootstrap_95_interval": [
-0.00717857607088874,
0.00584743262658923
],
"improved": 1,
"equal": 166,
"regressed": 3
},
{
"candidate": "r24-new-queries",
"baseline": "r20-t2ranking-zh8192",
"group": "relevant_document_above_8192_tokens",
"queries": 166,
"candidate_score": 0.845615412812933,
"baseline_score": 0.8471364337661693,
"delta": -0.001521020953236308,
"paired_bootstrap_95_interval": [
-0.007675908408223043,
0.004130748431944928
],
"improved": 1,
"equal": 160,
"regressed": 5
}
],
"r20_reference_content_sha256": "dbbba2a09857723e44f77f6084a22cbc5e49b6b281d95eaa545abbe546edde04",
"r20_reference_manifest_sha256": "99e03e6512f7d77defc4f43157822f3545742cd677824474f2bf2fcd3d3fafa6",
"runner_sha256": "5c7b45cbfb3619db1eff32b6232eb87a990eae6eb850670946f7f0e5607b20d6",
"launch_manifest_sha256": "7c2606f42cab34d637d58d3235dafc4b3290154d818ce4daddd3b078e4fe3688",
"lengths": {
"minimum": 2662,
"maximum": 23856,
"sum": 2887533,
"count": 336
},
"limitations": [
"Conditional development-query uncertainty; fixed corpus retained.",
"Intervals are exploratory and not adjusted for task/subgroup multiplicity.",
"A single long retrieval task does not establish all long-context or 32768-token endpoint capabilities."
]
},
"scope": "Previously exposed development validation, not fresh confirmation. Native FP32 MATH SDPA, not ONNX benchmark scores."
}