Dontopedia

spherical VQ

From Dontopedia, the open, paraconsistent wiki. (Last updated 2026-06-06.)

spherical VQ is vector quantization on spheres.

44 facts·27 predicates·15 sources·6 in dispute

Mostly:has performance concern(6), has good pattern(4), includes component(3)

Maturity scale raw canonical shape-checked rule-derived certified

Inbound mentions (14)

Other subjects in dontopedia point AT this entity as a value. These are inverse relationships — e.g. "X motherOf this subject" — and answer questions the forward facts can't. Grouped by predicate.

containsComponentContains Component(2)

advocatesReplacingAdvocates Replacing(1)

existentiallyCommittedToExistentially Committed to(1)

hasItemHas Item(1)

isIs(1)

isTargetFrameworkIs Target Framework(1)

lacksDocumentationForLacks Documentation for(1)

outputsToOutputs to(1)

precedesStagePrecedes Stage(1)

providesMlXPerformanceReviewForProvides ML X Performance Review for(1)

ranksAsSecondOptionRanks As Second Option(1)

stage3Stage3(1)

transformsIntoTransforms Into(1)

Other facts (40)

The long tail: predicates that appear too rarely to warrant their own section. Filter or scroll to find a specific one. Each row links to its source.

40 facts
PredicateValueRef
Has Performance ConcernQuantize Euclidean Z Sq Redundant[2]
Has Performance ConcernCompute Diagnostics Heavy[2]
Has Performance ConcernItem in Revive Dead Codes[2]
Has Performance ConcernL2 Normalize Duplicates Lohe Normalize[2]
Has Performance ConcernLog Norm Gain Mode No Op[2]
Has Performance ConcernUpdate Ema Allocates[2]
Has Good PatternL2 Normalize Uses Plus Eps Inside Sqrt[2]
Has Good PatternSampled Separation Penalty[2]
Has Good PatternSte Via Stop Gradient[2]
Has Good PatternVectorized Dead Code Revival[2]
Includes ComponentSphericalcodebook[7]
Includes ComponentSphericalvqbottleneck[7]
Includes ComponentSphericalvqhead[7]
Has ClassSpherical Codebook[15]
Has ClassSpherical Vq Bottleneck[15]
Has ClassSpherical Vq Head[15]
Rdf:typeArchitecture[10]
Rdf:typeSoftware Component[12]
Has Tests to WriteTests for Spherical Vq[1]
Relies on ML X PrimitivesPrimitives Py[2]
Transforms IntoStructural Codes[3]
May DuplicateLohe Dynamics Function[4]
Maintains Explicit CodebookExplicit Codebook[4]
Contrasts WithAnchor Kan[5]
Is Strong FallbackMechanism[5]
IsPhase 2 work[6]
Deontically Prioritizesif scaling fails[6]
Hypothesized to SolveBinding Issue[6]
Implicates Discreteness NecessityEntity Addresses[6]
Is Bestnext move if scaling fails[6]
PerformsVector Quantization on Spheres[7]
Includes Bottlenecktrue[8]
Includes Headtrue[8]
Includes Codebooktrue[8]
Precedes StageCode Lm[9]
Defined onS D 1 Surface[10]
Outputs toCode Lm[13]
Performs FunctionChoose Nearest Code[14]
Duplicates Function ofLohe Dynamics[14]
Descriptionvector quantization on spheres[15]

Timeline

Timeline axis is valid_time — when each source says the fact was true in the world, not when Dontopedia learned about it. Retracted rows are kept for provenance; coloured stripes indicate the context kind.

hasTestsToWriteblah/watt-activation/part-279
ex:tests-for-spherical-vq
hasPerformanceConcernblah/watt-activation/part-297
ex:quantize-euclidean-z-sq-redundant
hasPerformanceConcernblah/watt-activation/part-297
ex:compute-diagnostics-heavy
hasPerformanceConcernblah/watt-activation/part-297
ex:item-in-revive-dead-codes
hasPerformanceConcernblah/watt-activation/part-297
ex:l2-normalize-duplicates-lohe-normalize
hasPerformanceConcernblah/watt-activation/part-297
ex:log-norm-gain-mode-no-op
hasPerformanceConcernblah/watt-activation/part-297
ex:update-ema-allocates
reliesOnMlXPrimitivesblah/watt-activation/part-297
ex:primitives-py
hasGoodPatternblah/watt-activation/part-297
ex:l2-normalize-uses-plus-eps-inside-sqrt
hasGoodPatternblah/watt-activation/part-297
ex:sampled-separation-penalty
hasGoodPatternblah/watt-activation/part-297
ex:ste-via-stop-gradient
hasGoodPatternblah/watt-activation/part-297
ex:vectorized-dead-code-revival
transformsIntoblah/watt-activation/part-300
ex:structural-codes
mayDuplicateblah/watt-activation/part-309
ex:lohe-dynamics-function
maintainsExplicitCodebookblah/watt-activation/part-309
ex:explicit-codebook
contrastsWithblah/watt-activation/part-371
ex:anchor-kan
isStrongFallbackblah/watt-activation/part-371
ex:mechanism
isblah/watt-activation/part-370
Phase 2 work
deonticallyPrioritizesblah/watt-activation/part-370
if scaling fails
hypothesizedToSolveblah/watt-activation/part-370
ex:binding-issue
implicatesDiscretenessNecessityblah/watt-activation/part-370
ex:entity-addresses
isBestblah/watt-activation/part-370
next move if scaling fails
includesComponentblah/watt-activation/part-430
ex:sphericalcodebook
performsblah/watt-activation/part-430
ex:vector-quantization-on-spheres
includesComponentblah/watt-activation/part-430
ex:sphericalvqbottleneck
includesComponentblah/watt-activation/part-430
ex:sphericalvqhead
includesBottleneckblah/watt-activation/part-477
true
includesHeadblah/watt-activation/part-477
true
includesCodebookblah/watt-activation/part-477
true
precedesStageblah/watt-activation/part-302
ex:code-lm
typeblah/watt-activation/284
ex:Architecture
labelblah/watt-activation/284
Spherical VQ architecture
definedOnblah/watt-activation/284
ex:s-d-1-surface
labelblah/watt-activation/298
spherical VQ
typeblah/watt-activation/295
ex:SoftwareComponent
labelblah/watt-activation/295
Spherical VQ
outputsToblah/watt-activation/300
ex:code-lm
performsFunctionblah/watt-activation/307
ex:choose-nearest-code
duplicatesFunctionOfblah/watt-activation/307
ex:lohe-dynamics
labelblah/watt-activation/428
Spherical VQ
descriptionblah/watt-activation/428
vector quantization on spheres
hasClassblah/watt-activation/428
ex:spherical-codebook
hasClassblah/watt-activation/428
ex:spherical-vq-bottleneck
hasClassblah/watt-activation/428
ex:spherical-vq-head

References (15)

15 references
  1. [1]Part 2791 fact
    ctx:discord/blah/watt-activation/part-279
  2. [2]Part 29711 facts
    ctx:discord/blah/watt-activation/part-297
  3. [3]Part 3001 fact
    ctx:discord/blah/watt-activation/part-300
  4. [4]Part 3092 facts
    ctx:discord/blah/watt-activation/part-309
  5. [5]Part 3712 facts
    ctx:discord/blah/watt-activation/part-371
  6. [6]Part 3705 facts
    ctx:discord/blah/watt-activation/part-370
  7. [7]Part 4304 facts
    ctx:discord/blah/watt-activation/part-430
  8. [8]Part 4773 facts
    ctx:discord/blah/watt-activation/part-477
  9. [9]Part 3021 fact
    ctx:discord/blah/watt-activation/part-302
  10. [10]2843 facts
    ctx:discord/blah/watt-activation/284
    • full textwatt-activation-284
      text/plain2 KBdoc:agent/watt-activation-284/c0d87175-584d-4268-9b87-d373d011109e
      Show excerpt
      [2026-03-13 23:50] xenonfun: Claude: ``` ⏺ Clean takeaway: What works: - Spherical VQ architecture on S^{d-1} - Two-phase training (form codes, then use codes) - Assignment density scaling law (~4 pos/code minimum) - Direction+g
  11. [11]2981 fact
    ctx:discord/blah/watt-activation/298
    • full textwatt-activation-298
      text/plain2 KBdoc:agent/watt-activation-298/f5cde311-fd9a-43e7-a746-9177b5a91fee
      Show excerpt
      [2026-03-14 05:53] xenonfun: ``` What Changed The AR decoder produces recognizable English word fragments: "the", "and", "for", "with", "this", "one", "protec(tion)", "earch", "context", "project", "state", "imported". These are real m
  12. [12]2952 facts
    ctx:discord/blah/watt-activation/295
    • full textwatt-activation-295
      text/plain3 KBdoc:agent/watt-activation-295/3934680b-d58b-4c73-8470-2c337c1a045e
      Show excerpt
      [2026-03-14 04:39] xenonfun: ```❯ ⏺ Now I have the full picture. Here's my MLX performance review: Spherical VQ — MLX Performance Review Good patterns: 1. _l2_normalize uses + eps inside sqrt (line 38) — matches lohe_normalize sema
  13. [13]3001 fact
    ctx:discord/blah/watt-activation/300
    • full textwatt-activation-300
      text/plain3 KBdoc:agent/watt-activation-300/3b6edccf-3524-4608-838f-25890efaea15
      Show excerpt
      [2026-03-14 06:34] xenonfun: ``` 3. Manual attention (lines 110-128) — Hand-rolled softmax attention instead of using mx.fast.scaled_dot_product_attention. MLX's fused attention kernel is significantly faster for small sequence lengths.
  14. [14]3072 facts
    ctx:discord/blah/watt-activation/307
    • full textwatt-activation-307
      text/plain2 KBdoc:agent/watt-activation-307/bcc895dd-9148-4cd1-b67f-d61adcb12b77
      Show excerpt
      [2026-03-14 22:38] xenonfun: ```7. Why this matters for your architecture This is the part that is actually useful for you. Your pipeline is roughly: stream → harmonic lift → Lohe dynamics → readout stream→harmonic lift→Lohe dynamics→rea
  15. [15]4285 facts
    ctx:discord/blah/watt-activation/428
    • full textwatt-activation-428
      text/plain3 KBdoc:agent/watt-activation-428/2888b7ac-ac11-4017-8748-179fa76f2184
      Show excerpt
      [2026-03-20 02:56] xenonfun: ⏺ Significant gaps — 19 modules and 30 classes missing. The portal covers the high-level architecture but is missing: Major subsystems not documented: 1. Lohe Diffusion — image generation via Lohe manifold

See also

Keep researching

Missing something or suspicious of what's here? Kick off a research session — a Claude agent will investigate, cite its sources, and file new facts into a dedicated context you can review before accepting into the shared view.