github-actions[bot]
edddeee1ee
chore: Update NAPI-RS binaries for all platforms
...
Built from commit 00e8fc3f1e🤖 Generated by GitHub Actions
2026-02-08 17:07:51 +00:00
rUv
96c84f422e
feat: publish ruQu quantum simulation engine crates
...
Published crates:
- ruqu-core v2.0.2 - State-vector simulator
- ruqu-algorithms v2.0.2 - VQE, Grover, QAOA, Surface Code
- ruqu-exotic v2.0.2 - Quantum-classical hybrids
- ruqu-wasm v2.0.2 - WebAssembly bindings
Updated README with quantum engine section linking ADRs:
- QE-001 to QE-012: Core architecture to MinCut coherence
- Code example for GHZ state creation
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
2026-02-08 17:06:58 +00:00
github-actions[bot]
7970c07036
chore: Update NAPI-RS binaries for all platforms
...
Built from commit 8cfb2e8393🤖 Generated by GitHub Actions
2026-02-08 17:06:47 +00:00
rUv
efd0f70f56
Merge pull request #154 from ruvnet/claude/quantum-engine-adrs-6OsEO
...
feat: Add quantum simulation engine ADR series (QE-001 to QE-012) and DDD design documents
2026-02-08 12:05:03 -05:00
rUv
4a59e7e4de
Merge origin/main into claude/quantum-engine-adrs-6OsEO - resolve Cargo.toml conflict
2026-02-08 17:04:57 +00:00
rUv
7781e10c0e
docs: expand temporal tensor store section with PR #156 details
...
Added ADR links (018-023) and DDD reference for:
- Block-based storage engine
- Tiered quantization formats
- Temporal scoring tier migration
- Delta compression reconstruction
- WASM API cross-platform
- Benchmarking acceptance criteria
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
2026-02-08 17:03:22 +00:00
rUv
239f22214c
docs: update README with new crates and BitNet features
...
Added:
- ruvector-temporal-tensor: Temporal tensor store with tiered quantization
- ruvector-crv: CRV signal line protocol for vector search
- BitNet 1.58-bit quantization features to ruvllm description
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
2026-02-08 17:02:09 +00:00
rUv
2c607c7327
chore: bump versions for BitNet integration publish
...
- Workspace version: 2.0.1 → 2.0.2
- ruvector-sona: 0.1.4 → 0.1.5 (adds Debug impl for SonaEngine)
- ruvllm: 2.0.2 (BitNet integration from PR #151 )
Published crates:
- ruvector-sona v0.1.5
- ruvllm v2.0.2
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
2026-02-08 17:00:26 +00:00
github-actions[bot]
c0b610dcd1
chore: Update NAPI-RS binaries for all platforms
...
Built from commit 593b44a543🤖 Generated by GitHub Actions
2026-02-08 16:59:58 +00:00
github-actions[bot]
445dace141
chore: Update NAPI-RS binaries for all platforms
...
Built from commit e98ce26da3🤖 Generated by GitHub Actions
2026-02-08 16:58:30 +00:00
github-actions[bot]
926376e227
chore: Update NAPI-RS binaries for all platforms
...
Built from commit 0989957d15🤖 Generated by GitHub Actions
2026-02-08 16:55:50 +00:00
rUv
a213303f3a
Merge pull request #151 from ruvnet/claude/bitnet-ruvllm-research-Cz4Ot
...
docs: Add ADR-017 and DDD for Craftsman Ultra 30b 1bit BitNet integration
2026-02-08 11:55:15 -05:00
rUv
2c95c8085f
docs(ruvector-crv): add README for crates.io
...
Adds comprehensive README documenting the 6-stage CRV signal line
methodology mapping to ruvector subsystems. Bumped to v0.1.1.
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
2026-02-08 16:53:49 +00:00
rUv
a8e019f9f3
chore: add version specifications for crates.io publishing
...
Updated Cargo.toml files to specify explicit version requirements for
path dependencies, enabling successful publishing to crates.io.
Published crates:
- ruvector-temporal-tensor v2.0.1
- ruvector-core v2.0.1
- ruvector-gnn v2.0.1
- ruvector-raft v2.0.1
- ruvector-cluster v2.0.1
- ruvector-replication v2.0.1
- ruvector-graph v2.0.1
- ruvector-mincut v2.0.1
- ruvector-crv v0.1.0
- rvlite v0.3.0
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
2026-02-08 16:51:20 +00:00
github-actions[bot]
3d56868f28
chore: Update NAPI-RS binaries for all platforms
...
Built from commit 1f3e643e2c🤖 Generated by GitHub Actions
2026-02-08 16:44:32 +00:00
rUv
e9c8e11a6c
Merge pull request #156 from ruvnet/claude/temporal-tensor-compression-adr-Lc5Do
...
docs: Add ADR-018 through ADR-023 and DDD for temporal tensor store
2026-02-08 11:40:16 -05:00
Claude
f73f13c08a
feat: Add benchmarks for new features + persistence integration tests
...
Benchmarks (store.rs, 8 new bench tests):
- Batch scoring 10k blocks vs individual scoring
- 5-bit and 7-bit dequant fast paths (4096 values)
- 5-bit quantize fast path (4096 values)
- SVD adaptive rank selection (64x64 matrix)
- format_report and format_json throughput
- MetricsSeries trend computation (100 snapshots)
Persistence tests (10 tests, feature-gated):
- FileBlockIO: write/read, multi-tier, delete, overwrite, missing key
- FileMetaLog: append/get, upsert, iter, missing key, multi-block
354 tests pass (with --features persistence).
https://claude.ai/code/session_01Ksy165BL5nGpVoWaAfTE7t
2026-02-08 04:56:40 +00:00
Claude
8fa851c917
feat: Wire coherence gate, epoch tracker, and metrics series into TieredStore
...
- Add CoherenceCheck, EpochTracker, MetricsSeries fields to TieredStore
- put() now records write epochs for staleness detection
- tick() auto-records metrics snapshots for trend analysis
- Add enable_coherence()/disable_coherence() + accessor methods
- Add coherence_check() convenience method on TieredStore
- 4 new integration tests verify wiring
https://claude.ai/code/session_01Ksy165BL5nGpVoWaAfTE7t
2026-02-08 04:50:33 +00:00
Claude
ed8f369585
feat: Implement all 11 temporal tensor improvements
...
- 5-bit quantize fast path (8 values → 5 bytes, no bit accumulator)
- 5-bit + 7-bit dequant fast paths (8-at-a-time byte extraction)
- Batch scoring: compute_scores_batch, choose_tiers_batch,
score_and_partition, top_k_coldest with partial sort
- SVD improvements: reconstruction_error, energy_captured,
compression_ratio, from_data_adaptive (auto-rank selection)
- Metrics dashboard: format_report, format_json, health_check
with StoreHealthStatus, MetricsSeries with trend analysis
- Core trait integration (TensorStore, TensorStoreExt, TensorStoreSnapshot)
- AgentDB adapter (PatternIndex, InMemoryPatternIndex, AdaptiveTiering)
- Coherence gate (CoherenceCheck, EpochTracker, verify_put)
- Persistence layer (FileBlockIO, FileMetaLog, feature-gated)
- Stress/fuzz tests (8 adversarial scenarios)
- WASM FFI end-to-end test (feature-gated behind ffi)
306 tests pass (257 unit + 12 integration + 11 benchmarks +
14 property + 8 stress + 4 doctests).
https://claude.ai/code/session_01Ksy165BL5nGpVoWaAfTE7t
2026-02-08 04:34:12 +00:00
Claude
fb56bd45e1
perf: Optimize quantizer + store; add comprehensive tests
...
- Eliminate round() call in quantize hot path (1.8x speedup)
- Add 3-bit dequant fast path (8-values-from-3-bytes, 2.4x speedup)
- Wire WitnessLog into TieredStore (put/get/evict audit trail)
- Add TieredStore.metrics() for aggregate store statistics
- Add TieredStore.witness_log() accessors
- Update store.get() to accept `now` tick for access tracking
- 14 property-based tests (roundtrip, bitpack, segment, delta, f16,
score monotonicity, extreme values, compression ratio, determinism)
- 11 end-to-end integration tests (lifecycle, delta chain, quality
sweep, persistence, eviction, checksum, multi-tensor, stress,
compressor-to-store, factor reconstruction, witness logging)
Benchmarks (4096-element tensors, release mode):
8-bit quantize: 10,745 ns (1.52 GB/s)
8-bit dequant: 992 ns (16.52 GB/s)
3-bit dequant: 2,998 ns (5.46 GB/s)
Zipf P95 read: 41 ns
Tier flip rate: 0.074/block/min (threshold: 0.1)
All 204 tests pass.
https://claude.ai/code/session_01Ksy165BL5nGpVoWaAfTE7t
2026-02-08 03:57:57 +00:00
Claude
bc745b6115
feat: Implement temporal tensor store with block-based tiered compression
...
Implements the block-based storage engine specified in ADR-018 through
ADR-023 with 5 new modules and 1 benchmark/test suite.
New modules:
- store.rs (1056 lines): BlockKey, BlockMeta, Tier, TieredStore with
HashMap index, per-tier data storage, CRC32 checksums, eviction,
and BlockIO/MetaLog/Clock traits
- tiering.rs (846 lines): EMA + popcount + recency scoring with LUT-based
fast_exp_neg, hysteresis, min_residency, budgeted maintenance,
MigrationCandidate selection, warm aggressive mode (7->5 bit)
- delta.rs (825 lines): Sparse delta format (u16 index + i16 value),
DeltaChain with bounded length and compaction, FactorSet for
low-rank reconstruction, encode/decode serialization
- metrics.rs (770 lines): WitnessLog (ring buffer), WitnessEvent enum
(Access, TierChange, Eviction, Maintenance, Compaction, etc.),
StoreMetrics aggregates, StoreSnapshot serialization
- store_ffi.rs (680 lines): tts_init/put/get/tick/stats/touch/evict
WASM exports with u128 split into hi/lo u64, feature-gated
Optimizations:
- 8-bit fast path in quantizer: direct byte read/write, no bit
accumulator. Dequant: 7313ns -> 1290ns (5.7x faster, 12.7 GB/s)
- 8-bit fast path in bitpack: direct copy, no accumulator.
Pack: 8484ns -> 742ns (11.4x), Unpack: 8845ns -> 396ns (22.3x)
- #[inline] on hot functions
Benchmark results (release, 16KB blocks):
Quantize 8-bit: 18.9us Dequant 8-bit: 1.3us (12.7 GB/s)
Quantize 3-bit: 22.5us Dequant 3-bit: 7.2us (2.3 GB/s)
Score compute: 10ns Single frame decode: 178ns
Segment 8-bit decode: 1.5us (11.2 GB/s)
Zipf P95 read: 48ns Tier flip rate: 0.074/block/min
Quality (all PASS):
8-bit: 0.39% max error 7-bit: 0.79% max error
5-bit: 3.33% max error 3-bit: 16.67% max error
Tests: 170 unit + 12 integration/benchmark, all passing.
https://claude.ai/code/session_01Ksy165BL5nGpVoWaAfTE7t
2026-02-08 03:18:51 +00:00
Claude
5b1bf32ea7
docs: Add ADR-018 through ADR-023 and DDD for temporal tensor store
...
Extends ADR-017 with detailed architecture for block-based temporal
tensor compression. Introduces 6 new ADRs covering storage engine,
tiered quantization formats, temporal scoring algorithm, delta
compression/reconstruction, WASM API, and benchmarking criteria.
Adds comprehensive DDD with 5 bounded contexts mapping to the
proposed 6-crate workspace layout.
ADR-018: Block-based storage engine with BlockKey/BlockMeta model,
append-only MetaLog, tiered data files, CRC32 checksums
ADR-019: 8/7/5/3-bit quantization formats with two-level scale,
Tier0 compression-to-zero, codec_bits bit packing
ADR-020: EMA + popcount + recency scoring, hysteresis, budgeted
maintenance ticks, fast exp approximation
ADR-021: Read/write paths, sparse delta format, delta chain
compaction, factor-based reconstruction policies
ADR-022: WASM exports (tts_init/put/get/tick/stats), host-imported
BlockIO, cross-platform strategy (native/Node/browser/edge)
ADR-023: Zipf simulation acceptance test (1M blocks, 10M accesses),
microbench targets, failure mode catalog
DDD: 5 bounded contexts (Block Management, Quantization, Temporal
Scoring, Storage Engine, Delta & Reconstruction) with aggregate
roots, domain events, repository interfaces, and context map.
Total: 8,084 lines across 7 documents.
https://claude.ai/code/session_01Ksy165BL5nGpVoWaAfTE7t
2026-02-08 02:12:38 +00:00
Claude
aacbdec281
feat: Phase 2 cross-module discoveries — 6 new experiments, all validated
...
Discovery 5: Time-dependent disambiguation (decay + interference)
Faster-decohering meaning loses embedding structure over time,
shifting which meaning wins. "Financial" starts dominant but
"river" takes over as financial embedding decoheres faster.
Discovery 6: QEC on swarm reasoning chains (reasoning_qec + swarm)
Syndrome bits map to agent boundaries. Fired syndromes indicate
where adjacent agents disagree, enabling targeted identification
of incoherent reasoning steps.
Discovery 7: Counterfactual search explanation (collapse + reversible)
Removing each gate and measuring divergence reveals which operation
was most responsible for a search result. Ry gate (divergence=0.45)
vs identity-like gate (divergence=0.0).
Discovery 8: Syndrome-diagnosed swarm health (diagnosis + swarm)
Syndrome extraction localizes faults to the disruptor's neighborhood.
Low-health agent creates structural vulnerability that propagates
through connected components.
Discovery 9: Decoherence as differential privacy (decay + collapse)
Light noise (0.01): preserves top results, divergence=0.12, entropy=1.44
Heavy noise (1.0): randomizes results, divergence=0.61, entropy=2.07
Calibrated decoherence provides tunable privacy for embedding search.
Discovery 10: Full 4-module pipeline (decay→interfere→collapse→QEC)
Fresh knowledge (fidelity=0.99): correct results, 0 QEC syndromes
Stale knowledge (fidelity=0.28): corrupted results, QEC detects degradation
Pipeline degrades gracefully with automatic reliability signaling.
105 total tests: 57 lib + 42 Phase 1 integration + 6 Phase 2 discoveries
https://claude.ai/code/session_01B1NkbLDWYPaacS9miKsnvW
2026-02-06 16:25:02 +00:00
Claude
c0b4f90b7b
docs: Add ADR-QE-014 documenting exotic quantum-classical discoveries
...
Documents 4 validated Phase 1 discoveries:
- Decoherence trajectory fingerprinting (clustering without similarity)
- Interference-based polysemy resolution (microsecond disambiguation)
- Counterfactual dependency mapping (pipeline importance scoring)
- Phase-coherent swarm coordination (quality > headcount)
Outlines 8 Phase 2 hypotheses for cross-module experiments including
time-dependent disambiguation, QEC on swarm reasoning, counterfactual
search explanation, and the full decohere-interfere-collapse-verify pipeline.
https://claude.ai/code/session_01B1NkbLDWYPaacS9miKsnvW
2026-02-06 15:51:08 +00:00
Claude
c33ebf78f6
feat: Complete ruqu-exotic with all 8 modules, 99 tests passing, 4 discoveries
...
Add reversible_memory.rs: time-reversible quantum state with gate inversion,
rewind, counterfactual analysis, and sensitivity analysis.
Add reality_check.rs: browser-native verification circuits for superposition,
entanglement, interference, phase kickback, and no-cloning theorem.
Add comprehensive integration test suite (42 tests) covering all 8 exotic
modules plus 4 cross-module discovery experiments:
- Decoherence trajectory fingerprinting (similar embeddings decohere similarly)
- Interference-based polysemy resolution (context resolves word meanings)
- Counterfactual dependency mapping (identify critical vs redundant operations)
- Swarm phase alignment (phase-coherent agents outperform count-based voting)
Fix flaky unit tests in quantum_decay and quantum_collapse modules.
99 total tests: 57 lib + 42 integration, all passing.
https://claude.ai/code/session_01B1NkbLDWYPaacS9miKsnvW
2026-02-06 14:41:20 +00:00
Claude
b289e69032
feat: Add 4 exotic quantum-classical modules (decay, interference, QEC reasoning, syndrome diagnosis)
...
quantum_decay: Embeddings decohere via quantum noise channels instead
of TTL deletion. Phase fidelity degrades smoothly before magnitude,
giving structured forgetfulness.
interference_search: Concepts represented as amplitude superpositions.
Retrieval uses constructive/destructive interference in complex space
instead of cosine similarity reranking.
reasoning_qec: Reasoning steps encoded as qubits with repetition-code
syndrome extraction. Detects structural incoherence in reasoning traces
via parity checks between adjacent steps.
syndrome_diagnosis: System components mapped to a quantum graph.
Fault injection + syndrome extraction localizes fragile components
and identifies fault propagation paths.
https://claude.ai/code/session_01B1NkbLDWYPaacS9miKsnvW
2026-02-06 14:04:49 +00:00
Claude
99fdbf4a4c
feat(wip): Add ruqu-exotic crate scaffold with quantum collapse and swarm interference
...
Initial scaffold for 8 exotic quantum-classical hybrid algorithms:
- Quantum-shaped memory decay (embeddings decohere instead of deletion)
- Interference-based concept disambiguation (amplitude-space retrieval)
- Quantum-driven search collapse (superposition → measurement retrieval)
- Quantum-modulated agent swarms (interference instead of voting)
- Error-corrected reasoning traces (QEC on reasoning steps)
- Syndrome-based AI self diagnosis (fault localization via syndromes)
- Time-reversible memory (counterfactual debugging)
- Browser-native quantum reality checks (verification circuits)
Includes complete implementations for quantum_collapse and
swarm_interference modules. Remaining modules being implemented
by concurrent agents.
https://claude.ai/code/session_01B1NkbLDWYPaacS9miKsnvW
2026-02-06 02:26:33 +00:00
github-actions[bot]
586529e74f
chore: Update NAPI-RS binaries for all platforms
...
Built from commit c66dd1de5a🤖 Generated by GitHub Actions
2026-02-06 02:15:25 +00:00
rUv
354391f2fa
Merge pull request #153 from ruvnet/claude/temporal-tensor-compression-1r58N
2026-02-06 03:10:27 +01:00
Claude
2782fc6305
docs: Add ADR-QE-013 Deutsch's theorem proof with historical comparison
...
Complete proof of Deutsch's theorem with phase kickback lemma and
step-by-step derivation. Compares five major formulations:
- Deutsch (1985): original probabilistic version (p=1/2)
- Deutsch-Jozsa (1992): deterministic n-bit, 2 queries
- Cleve-Ekert-Macchiavello-Mosca (1998): deterministic, single query
- Nielsen-Chuang (2000): canonical textbook presentation
- Calude (2006): de-quantization using higher-dimensional classical bits
Includes de-quantization critique (Abbott et al.), classical wave
analogies, and analysis of when quantum advantage is genuine vs
artifactual.
Adds 6 verification tests to ruqu-algorithms confirming all four
oracles produce deterministic correct results via the ruqu-core
simulator, including a phase-kickback amplitude-level check.
https://claude.ai/code/session_01B1NkbLDWYPaacS9miKsnvW
2026-02-06 02:00:35 +00:00
Claude
d453d2da86
feat: Implement quantum simulation engine (ruqu-core, ruqu-algorithms, ruqu-wasm)
...
Full Rust implementation of the quantum simulation engine as specified
in ADR-QE-001 through ADR-QE-012:
ruqu-core: State-vector simulator with 2^n complex amplitudes, single
and two-qubit gate kernels (H, X, Y, Z, S, T, Rx, Ry, Rz, CNOT, CZ,
SWAP, Rzz), projective measurement with collapse, expectation values
for Pauli strings and Hamiltonians, gate fusion optimizer, circuit
builder API, and multi-shot simulator with noise model support.
ruqu-algorithms: VQE with hardware-efficient ansatz and parameter-shift
gradients, Grover's search with optimal iteration count, QAOA MaxCut
with Rzz phase separation, and distance-3 rotated surface code with
syndrome extraction and lookup decoder.
ruqu-wasm: WebAssembly bindings via wasm-bindgen exposing circuit
construction, simulation, Grover search, and QAOA to browser clients
with 25-qubit memory limit.
257 tests passing across all crates. Criterion benchmarks included for
gate throughput, bell state preparation, algorithm scaling, and memory
allocation across 4-20 qubit systems.
https://claude.ai/code/session_01B1NkbLDWYPaacS9miKsnvW
2026-02-06 01:24:14 +00:00
Claude
1ac58b38cf
docs: Add README for ruvector-temporal-tensor crate
...
Plain-language introduction explaining what temporal tensor compression
does and why it matters, feature tables, Quick Start with 4 code
examples (basic, streaming, random-access, custom policy), full API
reference, segment binary format spec, FFI/WASM guide, and build
instructions.
https://claude.ai/code/session_01U63xtGd5Q8mUevyY7nUSfJ
2026-02-06 01:01:48 +00:00
Claude
fbcc13c5f8
docs: Polish temporal tensor crate with clippy fixes, docs, and utilities
...
- Fix all clippy warnings: module-level //! docs, .div_ceil(), is_empty()
- Optimize segment::decode to call dequantize_f32 directly (skip legacy wrapper)
- Add decode_single_frame() for random-access frame decoding
- Add compression_ratio() utility for segment inspection
- Add comprehensive doc-examples with 3 tested examples in lib.rs
- Fix HEADER_SIZE offset bug in decode_single_frame (22 vs 26)
- All 41 unit tests + 3 doc-tests pass, 0 clippy warnings
https://claude.ai/code/session_01U63xtGd5Q8mUevyY7nUSfJ
2026-02-06 00:50:48 +00:00
Claude
b200b0cb47
perf: Optimize temporal tensor compression hot paths
...
Key optimizations:
- Eliminate per-value modulo/division in dequantize by restructuring
to iterate by frame→group→element (was: flat index with val_idx % tensor_len)
- Cache f32 scales in TemporalTensorCompressor to avoid repeated f16→f32
conversion on every push_frame (drift check + quantization)
- Add optimized _f32 API variants (frame_fits_scales_f32, quantize_and_pack_f32,
dequantize_f32) that accept pre-converted scales
- Pre-reserve Vec capacity in quantize_and_pack (avoids reallocations)
- Add #[inline] on qmax_from_bits, f32_to_f16_bits, f16_bits_to_f32
- Use chunks() iterator instead of manual index tracking in compute_scales
New tests (41 total, up from 33):
- Roundtrip tests for 5-bit and 7-bit quantization
- Non-finite value handling (NaN, Inf, -Inf)
- Single-element group edge case
- Compression ratio validation for all tiers
- Cold-tier (3-bit) full roundtrip
- Large tensor multi-group (512-dim, 50 frames)
- Accessor method coverage
https://claude.ai/code/session_01U63xtGd5Q8mUevyY7nUSfJ
2026-02-06 00:40:36 +00:00
Claude
501e02d753
feat: Add quantum simulation engine ADR series (QE-001 to QE-012) and DDD design documents
...
Comprehensive architecture decision records and domain-driven design documentation
for integrating a Rust-based quantum simulation engine (ruQu) into the ruVector stack.
ADR Series (12 documents):
- QE-001: Core Architecture - pure Rust state-vector simulator decision
- QE-002: Crate Structure - three-crate architecture (ruqu-core, ruqu-wasm, ruqu-algorithms)
- QE-003: WASM Compilation - WebAssembly strategy with 25-qubit limit enforcement
- QE-004: Performance Optimization - SIMD, multithreading, gate fusion, benchmarks
- QE-005: VQE Algorithm - variational eigensolver with exact expectation values
- QE-006: Grover Search - O(1) oracle optimization via direct state vector access
- QE-007: QAOA MaxCut - graph-based optimization with Rzz native gates
- QE-008: Surface Code Error Correction - mid-circuit measurement, syndrome extraction
- QE-009: Tensor Network Evaluation - MPS/contraction for shallow circuits
- QE-010: Observability & Monitoring - metrics, tracing, health checks integration
- QE-011: Memory Gating & Power Management - zero-idle, on-demand allocation
- QE-012: Min-Cut Coherence Integration - syndrome-to-decoder bridge with ruQu
DDD Design (3 documents):
- Strategic Design: 6 bounded contexts, context map, ubiquitous language
- Tactical Design: 6 aggregates, 20+ value objects, 15+ domain events, services
- Integration Patterns: anti-corruption layers, shared kernel, event flows
https://claude.ai/code/session_01B1NkbLDWYPaacS9miKsnvW
2026-02-06 00:39:39 +00:00
Claude
61d91a0ef9
feat: Add ADR-017 temporal tensor compression with tiered quantization
...
Introduces a complete temporal tensor compression system with:
- ADR-017: SOTA research-backed architecture decision record covering
groupwise symmetric quantization, temporal segment reuse, access-pattern
driven tier selection (8/7/5/3 bit), and WASM-compatible design
- ruvector-temporal-tensor crate (zero external dependencies):
- tier_policy: Score-based hot/warm/cold bit-width selection
- f16: Software IEEE 754 half-precision conversion
- bitpack: Arbitrary bit-width stream packing (no alignment waste)
- quantizer: Groupwise symmetric quantization with f16 scales
- segment: Binary segment format (TQTC) encode/decode
- compressor: Temporal segment manager with drift detection
- ffi: WASM/C FFI with handle-based resource management
- ruvector-temporal-tensor-wasm crate for wasm32 targets
- 33 passing unit tests covering all modules
Compression targets: 4x (hot/8-bit), 4.57x (warm/7-bit),
6.4x (warm/5-bit), 10.67x (cold/3-bit) vs f32 baseline.
https://claude.ai/code/session_01U63xtGd5Q8mUevyY7nUSfJ
2026-02-06 00:28:21 +00:00
Claude
30cd63958e
feat: Integrate ExpertPredictor prefetch, CompressedMlaCache, and E2E tests
...
- Wire ExpertPredictor into MoE forward path: predicts likely-next experts
from routing history and issues software prefetch hints (volatile read of
first cache line of predicted expert gate_proj weights) before routing runs
- Rebuild predictor every 16 tokens from routing history (amortized cost)
- Fix routing history tracking to target first MoE layer (config.first_k_dense_replace)
instead of hardcoded layer_idx==0 (layer 0 is Dense in GLM-4.7-Flash)
- Integrate CompressedMlaCache as configurable mode (set_compressed_kv):
stores only c_kv + k_pe (576 dims) instead of full K/V (10240 dims) per
position (~17.8x memory reduction), recomputing K_nope and V during attention
- Add mla_caches field initialized per-layer in load_gguf(), cleared in reset_cache()
- Add 13 new tests (216 total, all passing):
- E2E: forward produces logits, forward_token with KV cache, determinism,
different tokens give different logits, expert predictor builds from inference,
cache reset, compressed KV toggle, scratch pool allocation
- Benchmarks: forward_token throughput, TL1 GEMV dispatch, RMSNorm, softmax,
expert_forward performance
https://claude.ai/code/session_011nTcGcn49b8YKJRVoh4TaK
2026-02-04 07:43:37 +00:00
Claude
133792a51e
perf: Ultra-optimize BitNet inference backend with SIMD dispatch, fused SwiGLU, and zero-alloc paths
...
- Wire AVX2 TL1 GEMV SIMD dispatch into backend hot path via tl1_avx2 module
with scalar LUT fallback for non-x86_64 platforms
- Add ScratchPool with 17 pre-allocated FP32 buffers for zero-alloc forward pass
- Fuse SwiGLU gate+up projections with 4-wide unrolled loop and unsafe indexing
- Optimize RMSNorm with 4-way parallel accumulator and fused scale pass
- Optimize softmax with reciprocal multiply instead of per-element division
- Optimize fp32_matvec_transposed with 4-wide unrolled dot product
- Optimize GQA attention with 4-wide unrolled score computation and skip for
negligible weights
- Add routing history tracking via Mutex<Vec<Vec<usize>>> for expert prediction
(interior mutability preserves LlmBackend Send+Sync trait compatibility)
- Pre-allocate KV caches (512 positions) in load_gguf()
- Add tl1_gemv_into() for zero-allocation GEMV into caller-provided buffers
- All 203 bitnet tests pass
https://claude.ai/code/session_011nTcGcn49b8YKJRVoh4TaK
2026-02-04 07:12:49 +00:00
github-actions[bot]
2c8e8a4549
chore: Update NAPI-RS binaries for all platforms
...
Built from commit 987f822efc🤖 Generated by GitHub Actions
2026-02-04 07:01:21 +00:00
rUv
98d0494f26
Merge pull request #149 from ruvnet/claude/crv-ruvector-integration-Fbl4V
2026-02-04 12:26:44 +05:30
Claude
971c1f9bf3
chore: Update reasoning bank patterns cache
...
https://claude.ai/code/session_011nTcGcn49b8YKJRVoh4TaK
2026-02-04 05:54:40 +00:00
Claude
f7c36764e5
feat: Add streaming generation, predictive expert prefetcher, and compressed MLA KV cache
...
- Streaming generation API (generate_streaming) with per-token callback,
early stopping, and GenerationStats for throughput metrics
- ExpertPredictor: transition-matrix based predictor that learns from
routing history to predict next experts with Laplace smoothing
- CompressedMlaCache: stores compressed latents (c_kv + k_pe) instead
of full K/V, achieving ~17.8x memory reduction for GLM-4.7-Flash
- 15 new tests (203 total bitnet tests, all passing)
https://claude.ai/code/session_011nTcGcn49b8YKJRVoh4TaK
2026-02-04 05:53:56 +00:00
Claude
4ae1ad9182
feat: Add GLM-4.7-Flash GGUF tensor mapping, MLA attention, and model validation
...
- TensorNameMapper resolves both llama.cpp (blk.*) and HuggingFace (model.layers.*) naming
- MLA (Multi-Head Latent Attention) with low-rank Q/KV compression (DeepSeek-V2 style)
- Stacked 3D expert tensor support (ffn_gate_exps → per-expert slicing)
- Shared expert + dense layer-0 support (MoeWithShared/Dense/Moe layer types)
- Updated BitNetModelConfig defaults to match GLM-4.7-Flash architecture
- Tensor discovery and model validation harness for GGUF files
- 188 passing tests (14 new)
https://claude.ai/code/session_011nTcGcn49b8YKJRVoh4TaK
2026-02-03 18:00:17 +00:00
Claude
acfb352f40
feat: Add real attention, KV cache, RoPE, and tokenizer to BitNet backend
...
Resolves the three blocking gaps that prevented end-to-end inference:
1. **Real attention layer** (was pass-through placeholder):
- AttentionWeights struct with Q/K/V/O ternary projections
- GQA (Grouped Query Attention) with configurable num_heads / num_kv_heads
- Pre-computed RoPE cos/sin tables (apply_rope)
- Per-layer KV cache for autoregressive generation
- forward_token() for efficient single-token inference with cache
- forward_layer_cached() with full attention computation
- forward_layer_nocache() legacy path for backwards compatibility
2. **Tokenizer integration** (was raw bytes → token IDs):
- load_tokenizer_from_gguf() extracts vocab + merges from GGUF metadata
- Byte-level fallback tokenizer (260 tokens) when GGUF has no vocab
- TokenizerBridge implements crate-level Tokenizer trait
- tok() accessor for direct tokenizer access
3. **generate() uses tokenizer** (was returning [token_id] strings):
- Encodes prompt via BPE tokenizer before forward pass
- Decodes generated tokens back to text
- generate_cached() for KV-cached autoregressive generation
- get_embeddings() now uses tokenizer for text encoding
- reset_cache() to clear KV state between sequences
Tests: 174/174 bitnet tests pass (9 new: RoPE, KV cache, tokenizer roundtrip,
attention weights, byte-level fallback, cache operations)
https://claude.ai/code/session_011nTcGcn49b8YKJRVoh4TaK
2026-02-03 17:39:58 +00:00
Claude
c85ba8a498
feat: Add appliance-optimized RLM embedder (Pi 5 + STM32 offload)
...
Implements AD-25 appliance deployment optimizations for the RLM recursive
sentence transformer embedder targeting Raspberry Pi 5 + 7 STM32 coprocessors:
- Pi 5 config presets: pi5_optimized() (2-iter, 3-neighbor) and pi5_streaming() (1-iter)
- STM32 offload protocol: ComputeHash, FilterNeighbors, GateCheck, WatchdogPing, ScheduleReorder
- NullStm32 software fallback for development/cloud environments
- Batch embedding with per-chunk latency tracking and STM32 gate-checking
- Priority-scheduled batch embedding via STM32-driven reordering
- HashEmbedder: lightweight FNV-1a pseudo-embedder for testing/baseline
- FlatNeighborStore: in-memory neighbor retriever for small corpora (<100K chunks)
- EmbedderBenchmark: throughput, P95/P99 latency, peak memory reporting
- NEON-optimizable math: 4-element unrolled cosine_similarity, l2_normalize
- vec_accumulate_weighted and mean_embedding helpers
- 41 tests (27 new): STM32 protocol, batch, HashEmbedder, FlatNeighborStore, benchmark, integration
All 165 bitnet module tests pass.
https://claude.ai/code/session_011nTcGcn49b8YKJRVoh4TaK
2026-02-03 15:53:40 +00:00
Claude
767901ea79
feat: Add RLM embedder, tokenizer, eval gates, trace writer, and security hardening
...
New modules (4 files, 2,359 lines):
- rlm_embedder.rs (743L): RLM-style recursive sentence transformer with
3 variants (query-conditioned, corpus-conditioned, contradiction-aware
twin), merge rule, BaseEmbedder/NeighborRetriever traits, 14 tests
- tokenizer.rs (418L): BPE tokenizer with GGUF vocab loading, encode/decode,
special token handling, 10 tests
- trace.rs (554L): JSONL trace writer for routing, citation, refusal
decisions, jaccard similarity, manual JSON serialization, 10 tests
- eval.rs (644L): Three behavioral gates (routing correctness >= 0.85,
citation precision >= 0.90, refusal F1 >= 0.85), EvalSuite, 12 tests
Documentation:
- AD-24: RLM-Style Recursive Sentence Transformer Embedder — 3 variants,
merge rule, training strategy, evaluation criteria, appliance fit
- DDD v2.6: 8 new ubiquitous language terms, 4 new open questions (#31-34)
- 3 new positive consequences (#31-33) for RLM embeddings
Security hardening (across 6 existing files):
- Path traversal validation in GGUF export
- Division-by-zero epsilon guards in quantizer
- Bounds validation on public function inputs
- NaN-safe softmax with -inf handling
138 tests pass, 0 compilation errors.
Total bitnet module: 9,632 lines across 16 files.
https://claude.ai/code/session_011nTcGcn49b8YKJRVoh4TaK
2026-02-03 15:40:59 +00:00
Claude
11e1e3af95
feat: Add AD-23 Phase-1 distillation, expert cache, and DDD updates
...
AD-23: Phase-1 Distillation via External GPU Teacher Artifacts
- One-time GPU job produces behavioral artifacts (routing traces,
sparse logits, preference labels) — not trained weights
- CPU-only refinement: router repair, LoRA correction, EWC++, policy
optimization using teacher artifacts
- Acceptance criteria: 200-prompt suite, all 3 behavioral gates,
stability under 10% corpus perturbation
expert_cache.rs: MoE expert hot-set caching (new file)
- ExpertCache with LRU/LFU/Adaptive eviction policies
- MoeBatchScheduler: reorder token execution by expert for cache reuse
- Prefetcher trait for future platform-specific prefetch intrinsics
- 12 tests (92/92 bitnet tests pass)
DDD v2.5: 6 new ubiquitous language terms (Teacher Artifact, Behavioral
Distillation, Router Repair, Sparse Logits, Corpus Perturbation) and
4 new open questions (#27-30) for Phase-1 operability.
https://claude.ai/code/session_011nTcGcn49b8YKJRVoh4TaK
2026-02-03 15:12:33 +00:00
Claude
cf8daa60c3
docs: Add AD-22 evaluation infrastructure and behavioral gates
...
Defines three ship/no-ship gates:
- Gate 1: Routing correctness (>= 85% teacher agreement)
- Gate 2: Citation correctness (precision >= 90%, recall >= 70%)
- Gate 3: Refusal calibration (F1 >= 0.85)
Includes JSONL trace schema, auto-labeling strategy using RuVector
signals (redundancy, cluster disagreement, mincut fragility), and
go/no-go rule requiring all gates to pass on same prompt suite run.
https://claude.ai/code/session_011nTcGcn49b8YKJRVoh4TaK
2026-02-03 14:57:32 +00:00
Claude
58806480b3
fix: Polish AVX2 and WASM SIMD128 kernel variants
...
Agent refinements to tl1_avx2.rs and tl1_wasm.rs — cleanup
of unused imports and linter warnings.
https://claude.ai/code/session_011nTcGcn49b8YKJRVoh4TaK
2026-02-03 14:42:30 +00:00
Claude
395e214106
feat: Implement BitNet inference stack — TL1 kernel, backend, GGUF export, RLM refiner
...
Phase 0 + 0.5 implementation (4,283 lines across 6 new files):
- tl1_kernel.rs (879L): TL1 ternary GEMV with NEON SIMD + scalar fallback,
INT8 activation quantization (absmax), LUT generation, 17 tests
- backend.rs (1,179L): Full BitNetBackend implementing LlmBackend trait,
GGUF model loading, MoE router (softmax gate + top-K), expert FFN
(SwiGLU via TL1 GEMV), RMSNorm, embedding/LM head, 12 tests
- gguf_export.rs (662L): GGUF v3 writer for BITNET_T158, FP16 conversion,
model export with BitNet metadata, validation, 8 tests
- rlm_refiner.rs (696L): Phase 0.5 orchestrator wiring MicroLoRA + EWC++ +
GRPO + ContrastiveTrainer, SIMD-only mode (AD-20), checkpointing, 10 tests
- tl1_avx2.rs (414L): AVX2 SIMD kernel variant (x86_64 conditional)
- tl1_wasm.rs (453L): WASM SIMD128 kernel variant (wasm32 conditional)
All 72 bitnet tests pass. Fixed 2 pre-existing compilation errors in
autodetect.rs and kernels/mod.rs.
https://claude.ai/code/session_011nTcGcn49b8YKJRVoh4TaK
2026-02-03 14:34:37 +00:00
