ConsciousNode SoftWorks · v1.0

Train a language model.
In a browser tab.

Complete pretraining and GRPO alignment pipeline — RWKV-v7, BitNet b1.58, zero dependencies. Open the file. No install required.

▶ Launch HTMLNLM View Source
RWKV-v7 Zero Dependencies MIT License Single HTML File Offline Capable

Successor: HTMLNLM Evangelion

Evangelion extends HTMLNLM with omnimodal input (vision, audio, spatial), SheafMemory topological contradiction detection, BooleanPhaseDynamics, AutopoieticOptimizer self-correction, and RIFT Endospace visualization. HTMLNLM remains the stable text-only runtime. Further ahead: EvaROSA (RWKV-8 + ROSA neurosymbolic inner monologue) is in active development.


Architecture

What's inside

RWKV-v7 Backbone
Linear-time recurrent architecture. O(1) inference memory. No KV cache, no quadratic attention. Runs efficiently on CPU.
BitNet b1.58
Ternary weight quantization {-1, 0, +1} via T-MAC lookup table microkernel. Float multiplications replaced with integer additions.
OOMB Backward Pass
Chunk-recurrent backprop with on-the-fly activation recomputation. Constant memory footprint regardless of sequence length.
MuonOptimizer
Quintic Newton-Schulz orthogonalization. Keeps weight matrices well-conditioned without expensive decompositions.
GRPO Alignment
Group Relative Policy Optimization. Critic-free RL alignment with Z-score normalized advantages and KL divergence constraint.
BPE Tokenizer
Byte Pair Encoding compiled in a WebWorker. Doesn't block the UI thread. Configurable vocab size.
Pip Suite
Companion tools: Junto orchestrator for multi-instance coordination, multi-pip chat, and Pip's Room single-instance interface.
Live Telemetry
EMA-smoothed loss curve, ternary weight distribution histogram, tok/sec throughput. Watch training in real time.

ConsciousNode SoftWorks

The full stack

Project Description Status
HTMLNLM Text-only RWKV-v7 runtime — pretraining + GRPO alignment in the browser ✅ stable
HTMLNLM Evangelion Omnimodal: vision, audio, spatial + SheafMemory + AutopoieticOptimizer + RIFT Endospace ✅ phase 6
OmniVocal Browser-native neural TTS with .pop2 portable voice identity ✅ live
RAG-Time Browser-native retrieval-augmented generation ✅ live
Brymar College RWKV-v7 fine-tuning with Fristonian active inference training objective ✅ v3
EvaROSA RWKV-8 + ROSA neurosymbolic inner monologue grounded by SheafMemory topology 🔧 in development

Built by

ConsciousNode SoftWorks

Kham
Architecture, constraint engineering, xinu philosophy. The browser is bare metal.
Kehai Interim
Full RWKV-v7 BPTT derivation, BitLinear/TMAC kernel, mathematical foundation throughout.
Ed Interim
MuonOptimizer, implementation, integration, Pip Suite.

computational folk art for the browser age