63430 / logbook.json
stresearch-dev's picture
Update logbook: Repro: Do Transformers Need Three Projections? Systematic Study of QKV Variants
5f2a7a1 verified
Raw
History Blame Contribute Delete
4.48 kB
{
"schema_version": 2,
"title": "Repro: Do Transformers Need Three Projections? Systematic Study of QKV Variants",
"emoji": "🎯",
"space_id": "stresearch-dev/63430",
"paper": null,
"tags": [
"icml2026-repro",
"paper-XGxBnVqK63"
],
"updated_at": "2026-07-25T18:28:52+00:00",
"root": {
"slug": "index",
"title": "Repro: Do Transformers Need Three Projections? Systematic Study of QKV Variants",
"file": "pages/index.md",
"children": [
{
"slug": "summary",
"title": "Summary",
"file": "pages/summary/page.md",
"children": []
},
{
"slug": "c3-pad-fraction-measurement",
"title": "C3: Pad Fraction Measurement",
"file": "pages/c3-pad-fraction-measurement/page.md",
"children": []
},
{
"slug": "c1-absolute-perplexity-deflation",
"title": "C1: Absolute Perplexity Deflation",
"file": "pages/c1-absolute-perplexity-deflation/page.md",
"children": []
},
{
"slug": "c2-relative-gap-dilution-headline-falsification",
"title": "C2: Relative Gap Dilution (Headline Falsification)",
"file": "pages/c2-relative-gap-dilution-headline-falsification/page.md",
"children": []
},
{
"slug": "metric-provenance-tables-4-9-vs-table-11",
"title": "Metric Provenance: Tables 4/9 vs Table 11",
"file": "pages/metric-provenance-tables-4-9-vs-table-11/page.md",
"children": []
},
{
"slug": "claim-2-q-k-v-fully-shared-variant",
"title": "Claim 2: Q=K=V Fully-Shared Variant",
"file": "pages/claim-2-q-k-v-fully-shared-variant/page.md",
"children": []
},
{
"slug": "claim-3-q-k-v-mqa-gqa-4",
"title": "Claim 3: Q-K=V + MQA / + GQA-4",
"file": "pages/claim-3-q-k-v-mqa-gqa-4/page.md",
"children": []
},
{
"slug": "claim-5-k-v-cosine-similarity",
"title": "Claim 5: K/V Cosine Similarity",
"file": "pages/claim-5-k-v-cosine-similarity/page.md",
"children": []
},
{
"slug": "c1-c2-training-baseline-vs-q-k-v-scaled-down-cpu",
"title": "C1/C2: Training Baseline vs Q-K=V (scaled-down, CPU)",
"file": "pages/c1-c2-training-baseline-vs-q-k-v-scaled-down-cpu/page.md",
"children": []
},
{
"slug": "bug-report",
"title": "Bug Report",
"file": "pages/bug-report/page.md",
"children": []
},
{
"slug": "limitations",
"title": "Limitations",
"file": "pages/limitations/page.md",
"children": []
},
{
"slug": "corrected-claim-statement",
"title": "Corrected Claim Statement",
"file": "pages/corrected-claim-statement/page.md",
"children": []
}
]
},
"traces": [
{
"id": "88a401ce-4080-4d2e-b854-c79f17e8622c",
"title": "Falsification run: env setup through logbook draft",
"provider": "Claude Code",
"model": "claude-sonnet-5",
"started_at": "2026-07-23T17:15:13.695000+00:00",
"ended_at": "2026-07-25T18:28:00.096000+00:00",
"duration_ms": 177166401,
"event_count": 1573,
"turn_count": 45,
"source_available": true,
"attached_at": "2026-07-23T20:55:50+00:00",
"index_file": "traces/88a401ce-4080-4d2e-b854-c79f17e8622c/index.json"
}
],
"workspace": {
"file": "workspace.json",
"file_count": 16,
"total_size": 502329556,
"bucket_id": "stresearch-dev/63430-artifacts"
},
"agent_view_tokens": 18583,
"trace_view_tokens": 95021,
"workspace_view_tokens": 230,
"revision": "af41b8fded1d85f455e4",
"traces_ref": {
"repo_id": "stresearch-dev/63430-traces",
"repo_type": "dataset",
"repo_url": "https://huggingface.co/datasets/stresearch-dev/63430-traces",
"private": false,
"viewer_path": "trackio/index.json"
},
"trace_dataset": "https://huggingface.co/datasets/stresearch-dev/63430-traces",
"workspace_ref": {
"repo_id": "stresearch-dev/63430-artifacts",
"repo_type": "bucket",
"repo_url": "https://huggingface.co/buckets/stresearch-dev/63430-artifacts",
"private": false
},
"workspace_bucket": "https://huggingface.co/buckets/stresearch-dev/63430-artifacts"
}