context

Part 129

ctx:discord/blah/general/part-129

kind: custommode: permissive78 statements34 distinct subjects

No external document is attached to this context. (Many contexts are pure organisational labels.)

Facts in this context

Grouped by subject. Each subject links to its full article.

Ajaxdavis15 facts`ex:ajaxdavis`

attached	Avani Kelly Timeline Pdf
attached	Rosie and Owen Reynolds Year by Year Timeline 1 1 Pdf
attached	Lisa Watts Family Year by Year Timeline Pdf
commentsOnWorkEthic	Girvo
evaluatedAs	Fp4 Training Idea
expectsReportFrom	Girvo
expressesEnthusiasmFor	Fp4 Experiment
foundDepressing	Claude Oom Resolution
instructed	report back
posted	no sleep for the wicked
postedLink	Familysearch Ark 61903 1 1 Vqnx Gzn
postedMessageAt	2026-04-11 14:04
postedReplyAt	2026-04-11 12:36
sharesResources	Genealogy Timelines
watched	Claude

Girvo14 facts`ex:girvo`

addressedUser	User 806444151422976035
advocatesFor	Fp4 Native Training
attached	Image Png
believesPossible	Fp4 Model Training on Gb10
experimentsWith	Gb10
isKeenToSee	Gb10 Capabilities
isLookingToTrain	Fp4 Nearly Entirely 4 Bit Model Pipeline
posted	it begins...
postedMessageAt	2026-04-11 13:14
postedMessageAt	2026-04-11 12:22
postedMessageAt	2026-04-11 12:42
quotedText	The interesting question is whether you could train a small model...
signalsStartOf	Fp4 Training Experiment
wasNerdSnipedBy

Gb105 facts`ex:gb10`

canDo	1 petaflop
isBasicallyBuiltFor	Fp4
ofCalcsIn	Fp4
presupposesHighPerformance	Fp4 Calcs
supports	1 petaflop FP4 calcs

Claude3 facts`ex:claude`

didSoBy	jumping back and fourth
struggledWith	Precision Formats
triedToResolve	Oom Errors

Community Consensus3 facts`ex:community-consensus`

isNotQuestionedBy	very few people
isToCompressAfter	Bf16 Training
isToTrainBigIn	Bf16

Fp43 facts`ex:fp4`

addsLayerOf	Extra Fuckery
hasLogarithmicDistribution	[object Object]
soundsLike	Another Layer of Extra Fuckery

Chat Thread2 facts`ex:chat-thread`

mixesTopics	Family Timelines
mixesTopics	ML Training

Claude Oom Resolution2 facts`ex:claude-oom-resolution`

occurredBefore	Fp16 Fp32 Bf16
wasDepressingTo	Ajaxdavis

Fp4 Training2 facts`ex:fp4-training`

causesExtraFuckery	Oom Errors
deviatesFrom	Standard Path

Lisa Watts Family Year by Year Timeline Pdf2 facts`ex:lisa-watts-family-year-by-year-timeline-pdf`

concerns	Lisa Watts
isFamilyTimeline	Lisa Watts Family

Rosie and Owen Reynolds Year by Year Timeline 1 1 Pdf2 facts`ex:rosie-and-owen-reynolds-year-by-year-timeline-1-1-pdf`

concerns	Rosie Reynolds
concerns	Owen Reynolds

Small Model2 facts`ex:small-model`

isDesignedFor	Fp4
isTrainedFromGroundUp	Fp4

Training Small Fp4 Model From Ground Up2 facts`ex:training-small-fp4-model-from-ground-up`

challenges	Community Consensus
wouldProduce	Genuine Insights

Ajaxdavis Message 12 361 fact`ex:ajaxdavis-message-12-36`

repliesTo

Girvo Message 12 22

Ajaxdavis Message 14 041 fact`ex:ajaxdavis-message-14-04`

followsTemporally

Girvo Message 13 14

Architecture Choices1 fact`ex:architecture-choices`

playNicelyWith

Fp4 Logarithmic Distribution

Attachments1 fact`ex:attachments`

implyTopicShiftTo

Genealogy

Avani Kelly Timeline Pdf1 fact`ex:avani-kelly-timeline-pdf`

concerns

Avani Kelly

Bf161 fact`ex:bf16`

usedInBigModels

[object Object]

Bf16 Big Training Then Compress1 fact`ex:bf16-big-training-then-compress`

isAssumedOnlyViablePath

Community

Chat1 fact`ex:chat`

spansTime

2026-04-11 12:22 to 16:35

Community1 fact`ex:community`

holdsConsensusThat

Bf16 Then Compress

Familysearch Ark 61903 1 1 Vqnx Gzn1 fact`ex:familysearch-ark-61903-1-1-vqnx-gzn`

isGenealogyRecord

[object Object]

Fp161 fact`ex:fp16`

preceded

Bf16

Fp321 fact`ex:fp32`

preceded

Bf16

Fp4 Nearly Entirely 4 Bit Model Pipeline1 fact`ex:fp4-nearly-entirely-4-bit-model-pipeline`

isNearlyEntirely

4-bit

Genealogy Attachments1 fact`ex:genealogy-attachments`

suggestsResearchActivity

Ajaxdavis

Girvo Message 12 421 fact`ex:girvo-message-12-42`

repliesTo

Ajaxdavis Message 12 36

Girvo Message 13 141 fact`ex:girvo-message-13-14`

followsTemporally

Girvo Message 12 42

Loss Functions1 fact`ex:loss-functions`

penalizeDriftingInto

Fp4 Poorly Represented Regions

Nerd Sniping1 fact`ex:nerd-sniping`

presupposesInterestIn

Fp4 Training

No Sleep for the Wicked1 fact`ex:no-sleep-for-the-wicked`

alludesTo

Proverb

Quote Author1 fact`ex:quote-author`

posesQuestionAbout

Fp4 Native Training Possibility

Weight Initializations1 fact`ex:weight-initializations`

stayWithin

Fp4 Representable Range

Part 129

Facts in this context

Ajaxdavis15 factsex:ajaxdavis

Girvo14 factsex:girvo

Gb105 factsex:gb10

Claude3 factsex:claude

Community Consensus3 factsex:community-consensus

Fp43 factsex:fp4

Chat Thread2 factsex:chat-thread

Claude Oom Resolution2 factsex:claude-oom-resolution

Fp4 Training2 factsex:fp4-training

Lisa Watts Family Year by Year Timeline Pdf2 factsex:lisa-watts-family-year-by-year-timeline-pdf

Rosie and Owen Reynolds Year by Year Timeline 1 1 Pdf2 factsex:rosie-and-owen-reynolds-year-by-year-timeline-1-1-pdf

Small Model2 factsex:small-model

Training Small Fp4 Model From Ground Up2 factsex:training-small-fp4-model-from-ground-up

Ajaxdavis Message 12 361 factex:ajaxdavis-message-12-36

Ajaxdavis Message 14 041 factex:ajaxdavis-message-14-04

Architecture Choices1 factex:architecture-choices

Attachments1 factex:attachments

Avani Kelly Timeline Pdf1 factex:avani-kelly-timeline-pdf

Bf161 factex:bf16

Bf16 Big Training Then Compress1 factex:bf16-big-training-then-compress

Chat1 factex:chat

Community1 factex:community

Familysearch Ark 61903 1 1 Vqnx Gzn1 factex:familysearch-ark-61903-1-1-vqnx-gzn

Fp161 factex:fp16

Fp321 factex:fp32

Fp4 Nearly Entirely 4 Bit Model Pipeline1 factex:fp4-nearly-entirely-4-bit-model-pipeline

Genealogy Attachments1 factex:genealogy-attachments

Girvo Message 12 421 factex:girvo-message-12-42

Girvo Message 13 141 factex:girvo-message-13-14

Loss Functions1 factex:loss-functions

Nerd Sniping1 factex:nerd-sniping

No Sleep for the Wicked1 factex:no-sleep-for-the-wicked

Quote Author1 factex:quote-author

Weight Initializations1 factex:weight-initializations