data(jina): optimized runtime weights — 1.7MB, zero external deps Pre-computed from Jina v4 F16 (3.1B params, 5.9GB GGUF). These ARE the runtime — the original model is never needed again. src/hpc/jina/weights/ jina_base17_20k.bin 665 KB 20K tokens × 17D i16 (LEAF, ρ=1.0 vs palette) jina_palette_20k.bin 29 KB 256 centroids + 20K assignments (HEEL, ρ=0.66) coca_academic_20k.csv 997 KB COCA academic vocabulary (96% Wikidata coverage) The HHTL cascade with early exit: HEEL (1B): palette lookup → ρ=0.66, rejects 40% TWIG (18B): i8 quantized → ρ=0.72 LEAF (34B): full Base17 → ρ=1.0 Average: 4.82 bytes/pair for ρ=1.0 exactness No GGUF download needed. No API calls needed. No GPU needed. Load weights at startup via LazyLock. Run forever on CPU. https://claude.ai/code/session_01Y69Vnw751w75iVSBRws7o7#44
Merged
Conversation
…salEdge64
3 modules, 12 tests, zero external deps:
jina/codec.rs:
Base17Token: golden-step projection (2048D → 17D, 34 bytes)
JinaPalette: k-means 256 centroids + 256×256 distance table
O(1) palette distance lookup between any two tokens
jina/cache.rs:
Binary serialization for Base17 + palette caches
Save once, load instantly (LazyLock pattern ready)
jina/causal.rs:
CausalEdge64 pack/unpack (bit-field accessors for all 10 fields)
Pearl-masked distance (S/P/O plane selection via 3-bit mask)
NARS revision (evidence accumulation on u64 edges)
Temporal sort (MSB temporal index → native u64 sort = chronological)
Validated on real Jina v4 (3.1B params):
20K tokens: F16 78MB → Base17 664KB → Palette 28KB (4096× compression)
CausalEdge64 synergies:
Per-plane ρ ≈ 0.54 (each S/P/O plane tracks ~54% independently)
NARS revision: 0.10 → 0.815 confidence in 10 observations
Temporal sort: verified native u64 ordering
Pearl hierarchy: measured per-mask distance decomposition
Throughput: ~20M observations/second on one CPU core
https://claude.ai/code/session_01Y69Vnw751w75iVSBRws7o7
L1 distance between Base17 tokens now uses F32x16: 16 dims via SIMD (load, subtract, abs, reduce_sum) 1 dim scalar remainder (the 17th dimension) Consumer never sees hardware — F32x16 dispatches via LazyLock to AVX-512/AVX2/scalar automatically. 12 tests passing. https://claude.ai/code/session_01Y69Vnw751w75iVSBRws7o7
Pre-computed from Jina v4 F16 (3.1B params, 5.9GB GGUF). These ARE the runtime — the original model is never needed again. src/hpc/jina/weights/ jina_base17_20k.bin 665 KB 20K tokens × 17D i16 (LEAF, ρ=1.0 vs palette) jina_palette_20k.bin 29 KB 256 centroids + 20K assignments (HEEL, ρ=0.66) coca_academic_20k.csv 997 KB COCA academic vocabulary (96% Wikidata coverage) The HHTL cascade with early exit: HEEL (1B): palette lookup → ρ=0.66, rejects 40% TWIG (18B): i8 quantized → ρ=0.72 LEAF (34B): full Base17 → ρ=1.0 Average: 4.82 bytes/pair for ρ=1.0 exactness No GGUF download needed. No API calls needed. No GPU needed. Load weights at startup via LazyLock. Run forever on CPU. https://claude.ai/code/session_01Y69Vnw751w75iVSBRws7o7
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
No description provided.