Lohe Delta a

Achieves Bpb on Fineweb at StepsachievesBpbOnFinewebAtSteps

• 1.89^[1]all time · Part 696

Achieves Tokens Per SecondachievesTokensPerSecond

• 71000^[1]all time · Part 696

Comes With Synthetic Eval SuitecomesWithSyntheticEvalSuite

• Synthetic Eval Suite^[1]all time · Part 696

Good Fit forgoodFitFor

• moderate scope, concrete target^[1]all time · Part 696

Has Full Bptt inhasFullBpttIn

• Rust^[1]all time · Part 696

Has F Variant WithhasFVariantWith

• Clifford Binding Memory^[1]all time · Part 696

Has Total ParamshasTotalParams

• 28000^[1]all time · Part 696

Has VariantshasVariants

• 5 variants (A-E)^[1]all time · Part 696

Provides Concrete BenchmarksprovidesConcreteBenchmarks

• Synthetic Eval^[1]all time · Part 696

Requires Port Effort ofrequiresPortEffortOf

• ~3-5 days^[1]all time · Part 696

Similar to ModernsimilarToModern

• linear attention / state-space with delta-rule memory^[1]all time · Part 696

Trained for StepstrainedForSteps

• 20000^[1]all time · Part 696

Inbound mentions (4)

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.

Other facts (5)

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.

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.

References (2)

See also

• Synthetic Eval Suite
• Rust
• Clifford Binding Memory
• Audience
• Synthetic Eval
• Fineweb

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.