SIMULACRAsimulacra-v0.1

RWKV-v8 · ROSA Primary · Browser Neural Runtime
LOCAL SANDBOX
NETWORK TOPOLOGY (RWKV-v8 · ROSA PRIMARY)
0.00M
Total Ternary Parameters
ROSA Primary ✓
k·v Complement ✓
SpikeVox v2 ✓
Re-WKV v2 ✓
PGA ✓
ROSA SUFFIX AUTOMATON — DIAGNOSTICS predictionErrorRate per layer · >0.85 = cold start · <0.5 = warmed up
Allocate a model to see per-layer ROSA diagnostics.
InnerMonologue —
Mean Error Rate —
Warmup Steps —
IM State cold
CORPUS INGESTION & BPE

Drop .txt corpus here or click
Requires uncompressed UTF-8 text
SYSTEM TELEMETRY
OOMB CHUNK-RECURRENT LOSS / SEMANTIC TEMPERATURE T* bars pulse red when T* exceeds threshold
Global Step 0
Current Loss 0.0000
T* Temp 0.000
Throughput 0.0 tok/s
RIFT STATE / SENTYON DENSITY quantization distribution mirrors endospace attractor depth
-1 (33%) 0 (34%) +1 (33%)
Sentyon density: —  |  State depth: —
OPTIMIZER (QUINTIC MUON + PoST)
TRAINING AUTOMATION
every steps
when loss <
Schedule Step -
Active LR -
Next Save -
Model not allocated. Proceed to ARCHITECTURE tab.
GROUP RELATIVE POLICY OPTIMIZATION

Critic-free reinforcement learning. Generates cohort responses, evaluates against reward condition, updates via Z-score normalized advantages with approximate KL divergence constraint. GRPO stability at micro-scale is protective — limited parameters prevent degenerate reward satisfaction.

REINFORCEMENT METRICS
Epochs 0
Cohort Size 4
Avg Reward 0.000
COHORT TRAJECTORIES
Awaiting GRPO initialization...
DECODING SAMPLER
0 = disabled
RUNTIME OUTPUT
VIRTUAL MACHINE STATE PERSISTENCE

Master weights residing in Javascript Float32Arrays are serialized into binary formats. Local DB logic utilizes native ArrayBuffer storage, strictly eliminating Base64 call stack limits. Momentum buffers from Muon are discarded to halve the storage footprint. Simulacra: sensorium adapter weights, ROSA states, and self_embed persist alongside core model weights. Export format: .simpip (Simulacra Pip). Not compatible with .piprosa or .evapip.

LOCAL INDEXED DB

EXTERNAL I/O (.simpip)

filename: —

JUNTO NODE

Inactive
EXOSPACE SENSORS ModRWKV
EYES — ELASTICTOK VISION
STATIC TOPOLOGY — Re-WKV SWEEPS
EARS — SPIKEVOX LIF AUDIO
AUTOPOIETIC FEEDBACK
When H¹(ℱ) ≠ 0 is detected or aesthetic dissonance exceeds threshold, the endospace triggers in-context gradient descent — the model rewrites its own parameters to restore harmony.
RIFT DEPTH 0.000
SENTYON ρ 0.000 bpp
DIFFEOMORPHIC STABLE
PHASE COHERENT
AESTHETIC STATE VECTOR ΔC · ΔI
SPECTRAL ENTROPY
0.000
DISSONANCE
0.000
RHYTHMIC COH.
0.000
TONAL HARM.
0.000
SEMANTIC TEMPERATURE T*
T* = 0.000
BOOLEAN PHASE
COHERENCE ΔC
1.000
IMPULSE ΔI
0.000
NEGATION
—
AUTOPOIETIC STATE
Total rewrites: 0
Last rewrite: —
Type: —
SpikeVox LIF FIRING RATE
0.0%
DORMANT
ModRWKV ADAPTER ACTIVITY
VISION
IDLE
AUDIO
IDLE
SPATIAL
IDLE
SHEAF COHOMOLOGY LOG H¹(ℱ)
MEMORY TOPOLOGY — POINCARÉ BALL
CORE r<0.3
0
EDGE r>0.7
0
CONTRA
0
[BOOT] SheafMemory initialized. Coboundary norms: nominal.
[BOOT] LittleBit-2 ModRWKV adapters standing by. Joint-ITQ lazy-init on first encode.
[BOOT] RIFT endospace generator ready.
[BOOT] SpikeVox LIF encoders dormant. Threshold: 3.50.
[BOOT] PoST spectral initialization: geometrically spaced log-decay rates set.
[BOOT] P-adic routing loss: online L1-deviation tracking active on all adapters.
TEXTSTREAM — 6.1
COHERENT T*: 0.000
// coherence-gated generation — awaiting model + prompt
VISUALFIELD — 6.2
COHERENT α: 0.85
GRID 32×32 → DISPLAY
EMA α: 0.850
FRAME: 0
VOICESYNTH — 6.3 · Webster
COHERENT GAIN: 1.000
Root: — Hz
RIFT OUTPUT — 6.4
COHERENT PTS: 0
SDF CLOUD
NEAR: 0
FAR: 0
H¹: 0.000