Training System
From Dontopedia, the open, paraconsistent wiki. (Last updated 2026-06-06.)
Training System has 23 facts recorded in Dontopedia across 4 references, with 2 live disagreements.
Mostly:rdf:type(2), has feature(2), achieves speedup(1)
Maturity scale
raw canonical shape-checked rule-derived certifiedInbound mentions (3)
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.
announcesFeatureAnnounces Feature(1)
- Xenonfun
ex:xenonfun
describedFeatureDescribed Feature(1)
- Log Entry 2026 03 04 05 47
ex:log-entry-2026-03-04-05-47
mentionsTrainingSystemMentions Training System(1)
- Message 2026 03 15 16 45
ex:message-2026-03-15-16-45
Other facts (23)
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.
| Predicate | Value | Ref |
|---|---|---|
| Rdf:type | Software System | [3] |
| Rdf:type | Computational System | [4] |
| Has Feature | Parallel Tokenizer | [3] |
| Has Feature | Async Checkpoint Saving | [3] |
| Achieves Speedup | 68 | [1] |
| Has Async Checkpoint Saving | true | [1] |
| Has Parallel Tokenizer With Caching | true | [1] |
| Includes Philosophy Corpus Examples | true | [1] |
| Is Complete | true | [1] |
| Philosophy Corpus Dependent | Philosophy Corpus | [1] |
| Supports Dora Fine Tuning | true | [1] |
| Supports From Scratch Training With Best Checkpoints | true | [1] |
| Uses Learning Rate Scheduling | warmup + cosine/linear | [1] |
| Uses Mixed Precision Fp16 | true | [1] |
| Uses Optimized Mlx Conv1d | true | [1] |
| Involves Region | Fortage Region | [2] |
| Has Optimization | Mlx Conv1d Optimization | [3] |
| Supports Mode | From Scratch Training | [3] |
| Uses Learning Rate Schedule | Warmup Plus Schedule | [3] |
| Uses Precision | Fp16 | [3] |
| Uses Fine Tuning Method | Dora | [3] |
| Driven by | Packed Batch | [4] |
| Has State | Stable Regime | [4] |
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 (4)
ctx:discord/blah/watt-activation/part-18ctx:discord/blah/watt-activation/part-339ctx:discord/blah/watt-activation/18- full textwatt-activation-18text/plain3 KB
doc:agent/watt-activation-18/812b86dc-f424-4d37-b858-94cd90ca9fbfShow excerpt
[2026-03-04 05:47] xenonfun: feat: Complete training system with optimizations - 68x speedup with optimized MLX conv1d - Parallel tokenizer with caching - From-scratch training with best checkpoints - Async checkpoint saving - Learning rate…
ctx:discord/blah/watt-activation/249- full textwatt-activation-249text/plain2 KB
doc:agent/watt-activation-249/8436e2be-67b0-44b5-9c54-97c19c68045dShow excerpt
[2026-03-12 08:24] xenonfun: ⏺ Early packed phase data — only ~550 steps in but already showing interesting dynamics: r heatmap (top-left): Much more visible step-to-step fluctuation than the unpacked runs. The vertical striping pattern …
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.