KickModel
From Dontopedia, the open, paraconsistent wiki. (Last updated 2026-06-06.)
KickModel has 63 facts recorded in Dontopedia across 9 references, with 1 live disagreement.
Mostly:rdf:type(3), supports byte level(1), attempted generalization beyond(1)
Maturity scale
raw canonical shape-checked rule-derived certifiedInbound mentions (10)
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.
existsInKickmodelExists in Kickmodel(1)
- Freqweighted Eviction
ex:freqweighted-eviction
forModelFor Model(1)
- Triton Kernels
ex:triton-kernels
hasSubjectHas Subject(1)
- Comparison Wave Vs Kick
ex:comparison-wave-vs-kick
isComparableToIs Comparable to(1)
- Analytical V5 Run
ex:analytical-v5-run
isOptimalForIs Optimal for(1)
- Fibonacci S2 Quaternion Kick
ex:fibonacci-s2-quaternion-kick
lacksWordLevelCoherenceLacks Word Level Coherence(1)
- Ugt
ex:ugt
portedFromPorted From(1)
- Training Loop
ex:training-loop
presupposesExistenceOfPresupposes Existence of(1)
- Report
ex:report
sameAsInSame As in(1)
- Softmax Cross Entropy
ex:softmax-cross-entropy
selectsAsTopPickSelects As Top Pick(1)
- Xenonfun
ex:xenonfun
Other facts (60)
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 | Model Architecture | [7] |
| Rdf:type | Language Model | [8] |
| Rdf:type | Neural Network Architecture | [9] |
| Supports Byte Level | Byte Level | [1] |
| Attempted Generalization Beyond | Byte Level | [1] |
| Has Cache Entry Limit | 200 | [1] |
| Has Sparse Cache for | Bpe 8k | [1] |
| Uses Generation Method | Token-by-token | [2] |
| Achieves Speed | O(1)/token | [2] |
| Uses Cache Type | ConstellationCache | [2] |
| Knob Includes Momentum | Momentum | [3] |
| Knob Includes Lr | Learning Rate | [3] |
| Presupposes Autonomous Behavior After Kick | true | [3] |
| Activated by Kick | true | [3] |
| Underwent Analytical Training | Analytical Training Session | [3] |
| Has Limited Knobs | true | [3] |
| Part of Project | Harmonicmlx | [3] |
| Loaded Data From File | Fineweb Edu Bpe8k Npy | [3] |
| Loaded Data Bytes | 20480128 | [3] |
| Referenced in Comparison | Analytical V5 Run | [4] |
| Exists As Benchmark | True | [4] |
| Achieves Bpb | 2.54 | [4] |
| Uses | Quaternion Dynamics | [4] |
| Achieves Word Coherence | Word Level Coherence | [5] |
| Benchmark for Coherence | Word Level Coherence | [5] |
| Provides Training Infrastructure | True | [6] |
| Has Version | G8 Fibonacci | [6] |
| Keep Component | Fibonacci S2 Quaternion Kick | [6] |
| Baseline Model | True | [6] |
| Uses Loss | Softmax Cross Entropy | [6] |
| Has Mathematical Formulation | ∂ₜΨ = [Ω, Ψ] - γ⟨Ψ⟩ + kick | [7] |
| Described As | S³ oscillator | [7] |
| Uses Integration Method | exp-map geodesic integration | [7] |
| Uses Component | Omega Matrix | [7] |
| Has Variable | Psi State | [7] |
| Has Parameter | Gamma | [7] |
| Has Component | Kick Term | [7] |
| Sparse Extreme Mode | true | [9] |
| Recommended for | fast iteration | [9] |
| Total Parameter Count | 25 | [9] |
| Parameter Count Condition | G Equals 8 Configuration | [9] |
| Bits Per Byte | 2.04 | [9] |
| Evaluated on Dataset | Fineweb | [9] |
| Architecture Formula | ∂ₜΨ = [Ω, Ψ] − γ⟨Ψ⟩ + kick | [9] |
| Oscillator Type | S³ oscillators | [9] |
| Oscillator Parameter Count | 11 | [9] |
| Uses Exp Map Geodesic | true | [9] |
| Has Fifo Bivector Cache | true | [9] |
| Has Analytical Bptt | true | [9] |
| Analytical Bptt Speedup | 6.5 | [9] |
| Gradient Parity Check | Trivial Task | [9] |
| Fits in L1 Cache | true | [9] |
| Runtime Location | Gb10 Hardware | [9] |
| Potential Throughput | Millions of Steps in Minutes | [9] |
| Utility | Testing Optimization Tricks | [9] |
| Estimated Port Effort | ~1 day | [9] |
| Port Size | smallest port | [9] |
| Param Count Diagnostic Value | clean diagnostic | [9] |
| Gb10 Throughput Attribute | Hilarious | [9] |
| Role | sanity benchmark | [9] |
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 (9)
ctx:discord/blah/watt-activation/part-480ctx:discord/blah/watt-activation/part-486ctx:discord/blah/watt-activation/part-482ctx:discord/blah/watt-activation/part-495ctx:discord/blah/watt-activation/part-497ctx:discord/blah/watt-activation/part-496ctx:discord/blah/watt-activation/481- full textwatt-activation-481text/plain3 KB
doc:agent/watt-activation-481/00c8be1c-8839-4973-9fac-0dd3b65dedecShow excerpt
[2026-03-21 23:20] xenonfun: ``` ⏺ Here's the report for your expert: --- KickModel BPE-8K Results Report Architecture ∂ₜΨ = [Ω, Ψ] - γ⟨Ψ⟩ + kick — S³ oscillator with exp-map geodesic integration New Features Implemented 1.…
ctx:discord/blah/watt-activation/685- full textwatt-activation-685text/plain3 KB
doc:agent/watt-activation-685/d1abedb6-54a0-4109-8585-623e816ac07fShow excerpt
[2026-04-23 22:03] xenonfun: ``` /btw how might it apply in llm style like our other models or suppliment? Good question — a few concrete angles where CliffordNet's machinery could cross over to LLM-style use: Direct reuse of the …
ctx:discord/blah/watt-activation/693- full textwatt-activation-693text/plain3 KB
doc:agent/watt-activation-693/8372b985-fcfe-4f68-a730-4cd9d14a0fb9Show excerpt
[2026-04-28 23:22] xenonfun: KickModel (sparse extreme — top pick for fast iteration) 25 params total at g=8 (yes, twenty-five). BPB 2.04 on fineweb. Architecture: ∂ₜΨ = [Ω, Ψ] − γ⟨Ψ⟩ + kick, G independent S³ oscillators (3G+1 params), exp…
See also
- Byte Level
- Bpe 8k
- Momentum
- Learning Rate
- Analytical Training Session
- Harmonicmlx
- Fineweb Edu Bpe8k Npy
- Analytical V5 Run
- True
- Quaternion Dynamics
- Word Level Coherence
- G8 Fibonacci
- Fibonacci S2 Quaternion Kick
- Softmax Cross Entropy
- Model Architecture
- Omega Matrix
- Psi State
- Gamma
- Kick Term
- Language Model
- Neural Network Architecture
- G Equals 8 Configuration
- Fineweb
- Trivial Task
- Gb10 Hardware
- Millions of Steps in Minutes
- Testing Optimization Tricks
- Hilarious
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.