mx.compile
From Dontopedia, the open, paraconsistent wiki. (Last updated 2026-06-06.)
mx.compile has 27 facts recorded in Dontopedia across 12 references, with 2 live disagreements.
Mostly:rdf:type(2), potentially speeds up(2), applies to anchor more effectively(1)
Maturity scale
raw canonical shape-checked rule-derived certifiedInbound mentions (15)
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.
isCompatibleWithIs Compatible With(2)
- Phase 2
ex:phase-2 - Versor Phase 2
ex:versor-phase-2
lacksMxCompileLacks Mx Compile(2)
- Decode Step
ex:decode-step - Their Benchmark
ex:their-benchmark
benefitsSignificantlyMoreFromBenefits Significantly More From(1)
- Anchor Mode
ex:anchor-mode
breaksMxCompileBreaks Mx Compile(1)
- Resonance Attention Python Loop
ex:resonance-attention-python-loop
deniesUseOfDenies Use of(1)
- Xenonfun
ex:xenonfun
explainsBehaviorOfExplains Behavior of(1)
- Chat Message 3
ex:chat-message-3
forCompatibilityWithFor Compatibility With(1)
- Compiled Variant
ex:compiled-variant
incompatibleWithIncompatible With(1)
- Rotational Adamw
ex:rotational-adamw
isIncompatibleWithIs Incompatible With(1)
- Rotational Adam W
ex:rotational-adam-w
isMxCompileIs Mx Compile(1)
- Training Mode
ex:training-mode
needsFixForThroughputGapNeeds Fix for Throughput Gap(1)
- Lohe Spherical Plus Lohe V2
ex:lohe-spherical-plus-lohe-v2
noneUseNone Use(1)
- Smoketest Runs
ex:smoketest-runs
seesModestGainFromCompileSees Modest Gain From Compile(1)
- Pairwise Mode
ex:pairwise-mode
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 | Compilation Feature | [10] |
| Rdf:type | Compiler Feature | [12] |
| Potentially Speeds Up | Forward Pass | [10] |
| Potentially Speeds Up | Loss Computation | [10] |
| Applies to Anchor More Effectively | Anchor Mode | [1] |
| Fuses Kernels Well | Small Kernel Calls | [1] |
| Tested Without Layout Changes | Model Layout | [1] |
| References External Tool | Mx Framework | [1] |
| Causes No Divergence | Training Divergence | [1] |
| Potential Speedup Percent | 30-50 | [1] |
| Depends on Fixing | Tolist Issue | [2] |
| Was Identified Blocker | Tolist | [2] |
| Applies to Decode | Decode Step | [3] |
| Likely Closes Gap | null | [4] |
| Used With | Rotational Adamw | [5] |
| Optimizes Training | Training Run | [6] |
| Training Step Enabled | true | [6] |
| Is Python Mlx Feature | null | [7] |
| References Jax Tool | Jax | [8] |
| Bakes in Globals | Python Globals | [8] |
| Applies Speedup to | Rot Adamw Path | [10] |
| Does Not Affect | Forward Only Path | [10] |
| Would Help Performance of | Anchor Kan | [10] |
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 (12)
ctx:discord/blah/watt-activation/part-62ctx:discord/blah/watt-activation/part-106ctx:discord/blah/watt-activation/part-157ctx:discord/blah/watt-activation/part-203ctx:discord/blah/watt-activation/part-223ctx:discord/blah/watt-activation/part-238ctx:discord/blah/watt-activation/part-456ctx:discord/blah/watt-activation/part-275ctx:discord/blah/watt-activation/84- full textwatt-activation-84text/plain3 KB
doc:agent/watt-activation-84/16e41088-c84d-4a6f-9c2d-56d69830cfa6Show excerpt
[2026-03-07 20:41] xenonfun: okay some instant issues with this much data: ``` The problem: mx.eval(loss, model.parameters(), optimizer.state) traverses the full tree of 113M params + Adam's 2x state every step. For the compiled path, mx.ev…
ctx:discord/blah/watt-activation/220- full textwatt-activation-220text/plain3 KB
doc:agent/watt-activation-220/5c7f4a28-90e7-46de-ae1e-9e19a58c8d65Show excerpt
[2026-03-11 04:42] xenonfun: FFN DFT — much richer specialization than spectral: ``` ┌─────┬───────┬────────┬────────────────┬────────────────┐ │ blk │ r │ FFN DC │ dominant mode │ pattern │ ├─────┼───────┼────────┼───────…
ctx:discord/blah/watt-activation/222- full textwatt-activation-222text/plain3 KB
doc:agent/watt-activation-222/d8201f0f-b5d1-4b50-9f4e-2aca2c0d4c1eShow excerpt
[2026-03-11 05:02] xenonfun: ⏺ mx.compile with RotationalAdamW is a dead end — the optimizer creates new array objects on each step, so inputs=[model.state] captures stale references. The error "array without primitive" is exactly what CL…
ctx:discord/blah/watt-activation/447- full textwatt-activation-447text/plain2 KB
doc:agent/watt-activation-447/eb6b6320-13b6-4a43-9fc0-605d1212edf9Show excerpt
[2026-03-21 02:19] xenonfun: ``` ⏺ All phases complete. Here's the summary: Versor Extraction — Complete 20/20 tests pass, 686/686 full suite clean. What succeeded ┌───────────────────────────────┬───────────────────────────────…
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.