Lohe Spherical Plus Lohe V2

as donto believed on

From Dontopedia, the open, paraconsistent wiki. (Last updated 2026-05-02.)

Has Fewer Params Than Baseline	14M^[1]
Has Fewer Params Than Baseline	13000000^[2]
Has Global R	0.201^[1]
Has Global R	0.2053^[2]
Needs Fix for Throughput Gap	Grouped V Fix^[1]
Needs Fix for Throughput Gap	Mx Compile^[1]
Has Loss at Step	7.3735^[3]
Has Loss at Step	7.1846^[3]
Achieves Speed Percentage	67%^[1]
Expected Speed Percentage	85-90%^[1]
Has Attn R At1000	0.6915^[1]
Has Attn R Init	0.5845^[1]

ex:lohe-spherical-plus-lohe-v2

Lohe Spherical Plus Lohe V2 has 54 facts recorded in Dontopedia across 3 references, with 4 live disagreements.

54 facts·47 predicates·3 sources·4 in dispute

Mostly:has loss at step(5), has fewer params than baseline(2), has global r(2)

Has Fewer Params Than Baselinein dispute`hasFewerParamsThanBaseline`

14M^[1]all time · Part 203
13000000^[2]all time · Part 204

Has Global Rin dispute`hasGlobalR`

0.201^[1]all time · Part 203
0.2053^[2]all time · Part 204

Needs Fix for Throughput Gapin dispute`needsFixForThroughputGap`

Grouped V Fix^[1]all time · Part 203
Mx Compile^[1]all time · Part 203

Has Loss at Stepin dispute`hasLossAtStep`

7.3735^[3]all time · Part 201
7.1846^[3]all time · Part 201
7.3609^[3]all time · Part 201
7.3190^[3]all time · Part 201
7.4074^[3]all time · Part 201

Achieves Speed Percentage`achievesSpeedPercentage`

67%^[1]all time · Part 203

Expected Speed Percentage`expectedSpeedPercentage`

85-90%^[1]all time · Part 203

Has Attn R At1000`hasAttnRAt1000`

0.6915^[1]all time · Part 203

Has Global R At1000`hasGlobalRAt1000`

0.2008^[1]all time · Part 203

Has Mean Tok Per Sec`hasMeanTokPerSec`

8052^[1]all time · Part 203

Has Sys R At1000`hasSysRAt1000`

0.8817^[1]all time · Part 203

Has Throughput`hasThroughput`

8052^[1]all time · Part 203

Inbound mentions (13)

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.

`appliesToBothConfigs`Applies to Both Configs(1)

Mx Compile on Train Stepex:mx-compile-on-train-step

`causesThroughputIncrease`Causes Throughput Increase(1)

Grouped V Fixex:grouped-v-fix

`higherForLoheSpherical`Higher for Lohe Spherical(1)

Global Rex:global-r

`increasesDuringTraining`Increases During Training(1)

Sys Rex:sys-r

`indicatesCrossLayerPhaseCoherenceBuilding`Indicates Cross Layer Phase Coherence Building(1)

Sys R Risingex:sys-r-rising

`isComponentOf`Is Component of(1)

Lohe V2ex:lohe-v2

`isDoingRealWork`Is Doing Real Work(1)

Lohe Syncex:lohe-sync

`isValidated`Is Validated(1)

Grouped V Fixex:grouped-v-fix

`lowerForLoheSpherical`Lower for Lohe Spherical(1)

Wall Timeex:wall-time

`pretrains`Pretrains(1)

Test 1ex:test-1

`servesAsBaseline`Serves As Baseline(1)

Spectral Plus Lohe V2ex:spectral-plus-lohe-v2

`showsIncreasingSynchronization`Shows Increasing Synchronization(1)

Attn R Trajectoryex:attn-r-trajectory

`summarizesConfig`Summarizes Config(1)

Xenonfun Message 2026 03 10 06 54ex:xenonfun-message-2026-03-10-06-54

Other facts (35)

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.

35 facts

Predicate	Value	Ref
Has Train Loss Last50	7.851	[1]
Has Train Ppl Last50	2568.3	[1]
Has Val Loss Final	7.8019	[1]
Has Val Ppl	2445	[1]
Has Val Ppl Final	2445.1	[1]
Has Wall Time Mins	8.5	[1]
Has Wall Time Secs	508.7	[1]
Has Attn R	0.634	[2]
Has Fewer Params Percent Vs Baseline	12	[2]
Has Lower Param Count Than	Spectral Plus Lohe V2	[2]
Has Params	93766464	[2]
Has Slightly Worse Quality Than	Spectral Plus Lohe V2	[2]
Has Speed Multiplier Vs Baseline	1.23	[2]
Has Sys R	0.7049	[2]
Has Tok Per S	14371	[2]
Has Val Ppl	2565.3	[2]
Has Val Ppl Delta Vs Baseline	74.6	[2]
Has Wall Time Minutes	4.8	[2]
Has Worse Ppl by Percent	3	[2]
Is Compared to Baseline	Spectral Plus Lohe V2	[2]
Is Faster Than	Spectral Plus Lohe V2	[2]
Runs at Same Seq Batch As	Spectral Plus Lohe V2	[2]
Runs at Tok Per S	15000	[2]
Shows Stable Attn R	Attn R	[2]
With Grouped V Fix Is Percent of Spectral	128	[2]
Exhibits Loss Descent	Checkmark	[3]
Has Mean R Global	0.6116	[3]
Has R Global in Valid Range	Checkmark	[3]
Shows Loss Improvement	from 7.3609 to 7.1846	[3]
Improves Quality With Fewer Params	null	[1]
Commits to Phase Ontology	Attention Phases	[2]
Exists As Trained Model	Successful Training	[2]
Has Ppl Vs Baseline	2565	[2]
Preferred Over Baseline	Spectral Plus Lohe V2	[2]
Outperforms Loss Descent	Spectral Plus Lohe V2	[3]

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.

54 undated

—

achievesSpeedPercentageblah/watt-activation/part-203

67%

—

commitsToPhaseOntologyblah/watt-activation/part-204

ex:attention-phases

—

exhibitsLossDescentblah/watt-activation/part-201

ex:checkmark

—

existsAsTrainedModelblah/watt-activation/part-204

ex:successful-training

—

expectedSpeedPercentageblah/watt-activation/part-203

85-90%

—

hasAttnRblah/watt-activation/part-204

0.634

—

hasAttnRAt1000blah/watt-activation/part-203

0.6915

—

hasAttnRInitblah/watt-activation/part-203

0.5845

—

hasFewerParamsPercentVsBaselineblah/watt-activation/part-204

—

hasFewerParamsThanBaselineblah/watt-activation/part-203

14M

—

hasFewerParamsThanBaselineblah/watt-activation/part-204

13000000

—

hasGlobalRblah/watt-activation/part-203

0.201

—

hasGlobalRblah/watt-activation/part-204

0.2053

—

hasGlobalRAt1000blah/watt-activation/part-203

0.2008

—