feat(examples): universal_hhtld_encode — model-generic encoder with SlotL dispatch by AdaWorldAPI · Pull Request #183 · AdaWorldAPI/lance-graph

AdaWorldAPI · 2026-04-15T09:05:52Z

Summary

Step 4 of the universal_hhtld_encode plan (#179 mindset, #180 foundation, #181 HhtlDTensor, #182 SharedPaletteGroup).

Single example: crates/thinking-engine/examples/universal_hhtld_encode.rs (365 LOC). Model-generic — consumes any BPE-vocab safetensors model.

What it does

Load safetensors → bucket by (component, role, shape) → for each bucket:
    build_group_with_leaf(key, names, rows, k) which internally dispatches:
        argmax regime (qko/v/gate/up/down/projection) → 4 B/row Slot D only
        index  regime (embed/lm_head)                 → 12 B/row Slot D + Slot L
        passthrough   (norms/bias/<is_encodable)      → BF16 unchanged

Gates reported

#	Gate	Target	Measurement
1	Per-row ρ, argmax regime	median ≥ 0.95, p5 ≥ 0.90	sampled first 64 rows per tensor
1	Per-row ρ, index regime	median ≥ 0.98, p5 ≥ 0.95	same
3	Storage ratio vs BF16	≥ 2 : 1	total output bytes / orig BF16

Gates 2 (argmax-parity on held-out prompt) and 4 (WAV envelope vs raw) need integration with tts_full_inference.rs and land in a follow-up PR.

Usage

cargo run --release --example universal_hhtld_encode \
    --manifest-path crates/thinking-engine/Cargo.toml \
    -- /path/to/model.safetensors

Design notes

reconstruct_row_from_group rebuilds a transient HhtlDTensor from the SharedPaletteGroup's (cache, entries, slot_l, svd_basis) so it can call HhtlDTensor::reconstruct_row. Cleaner: add this as a method on SharedPaletteGroup itself — deferred to keep this PR focused on the validation example.
Sample cap of 64 rows per tensor bounds wall time on 151K-row vocab tensors. Gate 2 (argmax parity) requires the full row set and is handled in the follow-up integration PR.
bucket_tensors sorts results for stable reporting output across runs.

Stack

PR	Status	What
`#180`	merged	SlotL foundation module
`#181`	merged	HhtlDTensor × SlotL (`encode_with_leaf` / `reconstruct_row`)
`#182`	open — required	SharedPaletteGroup × SlotL (`build_group_with_leaf`)
`#183`	this PR — depends on #182	universal encoder example + gates 1 + 3

Test plan

cargo build --release --example universal_hhtld_encode — clean first try
cargo run on Qwen3-TTS-0.6B — currently running in background, full result will be posted as a comment when complete
Integrate gates 2 + 4 via tts_full_inference in follow-up PR

Follow-ups if this PR passes

Gates 2 + 4 PR — fork of tts_full_inference that loads weights via the encoded .hhtld → run raw vs encoded codec-token comparison + WAV envelope check
Disk serialisation — .hhtld container with magic byte header (only needed once gates pass end-to-end)
Model-family probes — run on Jina v5, Qwen3-SLM, Qwen3-VL (if open) to confirm model-generic claim

Follow-ups if GATES FAIL

Per the other session's ranked fallbacks:

Fail mode	First response
index ρ median < 0.98	Repack SlotL as `sign + i4
global SVD tops out < 0.99	Per-leaf local SVD basis (256 × 16 × n_cols overhead) per E5
argmax p5 < 0.90	Expand palette (k=256→512 or 1024) or add Slot V use beyond magnitude
ratio < 2:1	Audit passthrough count; tighten `is_encodable` threshold

https://claude.ai/code/session_01NYGrxVopyszZYgLBxe4hgj

Step 3 of the universal_hhtld_encode plan (#179 mindset doc, #180 SlotL foundation, #181 HhtlDTensor integration). The SVD basis for Slot L now lives at GROUP granularity — one basis amortised across all same-role same-shape tensors in the group. ## Changes (additive, backwards-compat) SharedPaletteGroup gains: - pub tensor_slot_l: Vec<(String, Vec<SlotL>, f32)> // per-tensor leaves - pub svd_basis: Option<SvdBasis> // shared basis Existing `build_group_with_fisher_z` constructor defaults both to empty / None. No caller needs to change. ## New dispatch primitive pub fn should_use_leaf(role: &str) -> bool true -> "embed" | "lm_head" (index-regime, per-row identity needed) false -> everything else (argmax-regime, 4 B/row is enough) Maps directly to the two-regime split named in docs/COMPRESSION_MINDSET_SHIFTS.md § "The insight that reframes the rest". ## New entry point pub fn build_group_with_leaf(key, names, rows_f32, k) -> SharedPaletteGroup Dispatches on key.role: - argmax-regime -> delegates to build_group_with_fisher_z (unchanged) - index-regime -> builds ONE SvdBasis from first tensor's rows (capped at 4096 sample rows for speed on 151K-vocab), then encodes each tensor via encode_with_leaf so the basis is shared across the whole group Wire cost per row for index-regime groups: 4 B (Slot D + Slot V) + 8 B (Slot L) = 12 B/row. Basis cost is amortised: one SvdBasis per group, regardless of tensor count. ## Convenience methods on SharedPaletteGroup slot_l_byte_size() -> bytes across all per-tensor Slot L entries svd_basis_byte_size() -> bytes for the shared SVD basis (0 if None) slot_l_for(tensor_name) -> Option<(&[SlotL], f32)> for lookup in reconstruction paths ## Tests (all new pass) should_use_leaf_classification ................................. ok build_group_with_leaf_falls_back_for_argmax_regime ............. ok role="qko" -> no SVD basis, no Slot L (4 B/row preserved) build_group_with_leaf_populates_slot_l_for_index_regime ........ ok role="embed", 2 tensors × 64 rows × 128 cols -> Slot L populated at 8 B/row, basis shared (single SvdBasis) svd_basis_shared_across_group_not_per_tensor ................... ok Confirms amortisation: basis_size is constant; entries_size scales linearly with tensor count. Full bgz-tensor suite: 150 passing, 4 new = 154. Pre-existing failures on main (gamma_calibration, hhtl_d_entry_roundtrip, matryoshka, hhtl_cache) are unchanged — not introduced by this work. ## Relation to prior PRs in session #180 (merged) - SlotL module (8 × i8 on shared SVD basis) #181 (merged) - HhtlDTensor × SlotL per-tensor integration #182 (this) - SharedPaletteGroup × SlotL group-level amortisation ## Follow-ups - `universal_hhtld_encode.rs` example: iterate over tensors, bucket by (classify_component, classify_role, effective_shape), feed each bucket to build_group_with_leaf (which internally dispatches on role) - .hhtld container format: single-file pack with magic byte header, palette + basis + per-tensor entries + Slot L - Inference wiring: swap tts_full_inference's RVQ codebook sum for HhtlDTensor::reconstruct_row on the index-regime tensors https://claude.ai/code/session_01NYGrxVopyszZYgLBxe4hgj

… with SlotL dispatch Implements the `universal_hhtld_encode` proposal from docs/COMPRESSION_MINDSET_SHIFTS.md (#179), built on the #180, #181, #182 foundation stack. ## What it does Consumes any BPE-vocab safetensors model, buckets tensors by (component, role, shape), routes each bucket through bgz_tensor::shared_palette::build_group_with_leaf which auto-dispatches on role: argmax regime (qko/v/gate/up/down/projection) → 4 B/row Slot D only index regime (embed/lm_head) → 12 B/row Slot D + Slot L passthrough (norms, biases, < is_encodable) → BF16 unchanged ## Validation gates This ships gates 1 + 3 of the 4-gate plan: GATE 1: per-row ρ histogram, split by regime - argmax: target median ≥ 0.95, p5 ≥ 0.90 - index: target median ≥ 0.98, p5 ≥ 0.95 GATE 3: storage ratio vs BF16 original - target ≥ 2:1 Sample-based (first 64 rows per tensor) to keep wall time bounded on 151K-row vocab tensors. Gates 2 (argmax-parity on held-out prompt) and 4 (WAV envelope match vs raw) require integration with tts_full_inference.rs and land in a follow-up PR. ## Usage cargo run --release --example universal_hhtld_encode \ --manifest-path crates/thinking-engine/Cargo.toml \ -- /path/to/model.safetensors ## Design notes - reconstruct_row_from_group rebuilds a transient HhtlDTensor from the SharedPaletteGroup's (cache, entries, slot_l, svd_basis) so it can call HhtlDTensor::reconstruct_row. Cleaner would be a method on SharedPaletteGroup; deferred to keep the PR focused on the example. - Sample cap at 64 rows per tensor: full validation pass on Qwen3-TTS-0.6B is O(bucket_count × tensor_count × 64 × n_cols) which bounds wall time. For gate 2 (argmax parity) the full row set matters — handled in the follow-up integration PR. ## Session PR stack #180 (merged) - SlotL foundation #181 (merged) - HhtlDTensor × SlotL per-tensor integration #182 (this PR's dep) - SharedPaletteGroup × SlotL group-level integration #183 (this PR) - universal_hhtld_encode example + gates 1 + 3 https://claude.ai/code/session_01NYGrxVopyszZYgLBxe4hgj

AdaWorldAPI · 2026-04-15T09:09:20Z

Full run results: ratio passed, reconstruction gates failed

Qwen3-TTS-0.6B end-to-end. Exit 0. ~5 min total wall.

ARGMAX regime: mean ρ = 0.040,  p50 = 0.033,  p5 = 0.000    FAIL
INDEX  regime: mean ρ = 0.157,  p50 = 0.090,  p5 = 0.019    FAIL
RATIO:         243.4 : 1   (1829.3 MB → 7.52 MB)             PASS

Better ratio than projected 126:1, but reconstruction is nearly zero correlation on both regimes.

Root cause

The reconstruction path uses Base17::to_f32(n_cols) to expand palette centroids. That function places 17 i16 values at GOLDEN_POS indices per 17-dim octave — other dimensions = 0. For a 1024-dim row, only ~60 of 1024 dims have non-zero centroid content before SlotL runs. SlotL then tries to correct a residual where the centroid offset is 94% zeros.

projection.rs and BGZ_HHTL_D.md both explicitly state this: "encoding is designed for the HHTL cascade, not for decompression back to f32."

I built SlotL-style reconstruction on a primitive that was never meant to decode. Same class of error as HCLAM 256×256 (PR #177 → refuted in #178): accepting a documented-but-unverified quality claim and building on top of it.

What this DOES confirm

HhtlDTensor is a lookup address, not a reconstruction centroid. Using it for f32 GEMM was the wrong application. Using it for HHTL cascade table-walks (the thing it was designed for) still works at 343:1.
encode_with_leaf algorithm is correct — the SlotL module test at feat(bgz-tensor): add SlotL — 8 × i8 leaf residual on shared SVD basis #180 passed with ρ ≥ 0.98 on synthetic data with zero-centroid baseline, isolating the SlotL arithmetic. The Base17 centroid expansion is the broken step, not SlotL itself.
The ratio is real — 243:1 stands regardless of reconstruction quality. The storage story works; the f32-reconstruction story does not.

Path A — f32 centroid palette (reconstruction-grade)

Parallel codec, new module crates/bgz-tensor/src/hhtl_f32.rs. Replace Base17 palette entries with CLAM centroids stored as f32 or BF16 vectors (256 × n_cols × 2–4 B = 1–2 MB per palette × 26 groups = 26–52 MB). Keep SlotL on top.

Projected ratio: 10–30:1
Projected ρ: median ≥ 0.95 argmax, ≥ 0.98 index
Scope: ~400 LOC new module, reuses SlotL + SvdBasis. ~1 session.

Path B — inference in codec space (what HhtlDTensor is for)

Swap tts_full_inference's forward pass from hidden @ reconstructed_W.T → argmax to HHTL cascade lookups via bgz_tensor::hhtl_cache (Skip/Attend/Compose/Escalate) + FisherZTable.

Ratio stays at 243:1 / published 343:1
Inference is a different computation, not f32 GEMM
Scope: multi-session. E5 of the mindset doc.
This is what HhtlDTensor was designed for; the current session kept trying to make it do something else.

Recommendation

Path A for the next PR. It's the one-session move that would land a shippable reconstruction-grade codec. Path B is the correct long-term architecture but a bigger rewrite.

Do NOT merge #183. It proves the failure mode but the example as it stands is misleading (243:1 ratio claim without the quality context). Either:

Add a prominent REFUTED banner + keep as historical artifact
Close feat(examples): universal_hhtld_encode — model-generic encoder with SlotL dispatch #183, rework as universal_hhtld_f32_encode.rs in a new PR using the Path A codec

Leaning toward (2) — keep history clean. #180, #181, #182 remain useful primitives that the new codec will consume.

Stack after this finding:

#178 open — passthrough fix + Lance roadmap + WAV test
#180 merged — SlotL foundation (still correct)
#181 merged — HhtlDTensor × SlotL (correct as an algorithmic building block, wrong as a reconstruction path)
#182 open — SharedPaletteGroup × SlotL (same caveat)
#183 this PR — proposed close in favor of Path A follow-up

https://claude.ai/code/session_01NYGrxVopyszZYgLBxe4hgj

Generated by Claude Code

chatgpt-codex-connector

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: d1d6d293a4

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

Open a pull request for review
Mark a draft as ready
Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

chatgpt-codex-connector · 2026-04-15T09:10:20Z

+    let total_output = total_entries_bytes + total_slot_l_bytes + total_svd_basis_bytes
+                     + total_palette_bytes + passthrough_bytes;


Include Fisher-Z bytes in total output accounting

Gate 3 currently computes total_output from entries, Slot L, SVD basis, palettes, and passthrough tensors, but omits the Fisher-Z tables that are built for every group (group.fisher_z). At k=256, each omitted table is about 64 KB, so across many groups this can undercount output by multiple MB and incorrectly report a PASS on the >= 2:1 storage gate. Please add Fisher-Z byte size to the per-group accumulation and final ratio/reporting.

Useful? React with 👍 / 👎.

chatgpt-codex-connector · 2026-04-15T09:10:20Z

+    if tensor_rows_f32.is_empty() {
+        return build_group_with_fisher_z(key, tensor_names, tensor_rows_f32, k);


Avoid panic when leaf builder receives no tensors

This empty-input guard still forwards to build_group_with_fisher_z, which immediately indexes tensor_rows_f32[0]. If build_group_with_leaf is called with an empty slice (for example after upstream filtering), this path will panic instead of handling the empty case. The guard should return a safe empty result (or explicit error) rather than dispatching to a function that assumes at least one tensor.

Useful? React with 👍 / 👎.

…#183) Per codex P1 comment: the per-group Fisher-Z table (k*k i8 + 8 bytes family gamma = ~64 KB at k=256) was omitted from the total output byte count. Across 26 groups on Qwen3-TTS-0.6B that's ~1.7 MB understatement — enough to paint a falsely generous ratio on marginal models and to undercount on all models. Changes: - total_fisher_z_bytes accumulator - group.fisher_z.as_ref().map(|f| f.byte_size()).unwrap_or(0) per group - included in total_output - new line in Gate 3 summary: "Fisher-Z tables: {mb} MB" Impact on the current run: total_output goes from 7.52 MB to ~9.2 MB (adds 1.66 MB), so the reported 243:1 ratio is closer to ~199:1. Ratio still passes the ≥ 2:1 gate by two orders of magnitude — the broken gate here is reconstruction ρ (documented in the PR #183 comment thread), not storage. This fix is useful even if the recommendation to close #183 stands: the accounting should be correct in any future Path A / Path B successor codec example. Refs: PR #183 codex comment (P1 badge, "Include Fisher-Z bytes in total output accounting") https://claude.ai/code/session_01NYGrxVopyszZYgLBxe4hgj

Post-#183 finding: Base17 palette substrate can't reconstruct rows for f32 GEMM (per-row ρ ≈ 0.04 on real Qwen3). This lands both paths the other session ranked as viable forward directions. ## Path A — HhtlF32Tensor (reconstruction-grade) New module crates/bgz-tensor/src/hhtl_f32.rs (5 tests passing): pub struct HhtlF32Entry { pub twig: u8 } // 1 byte/row pub struct HhtlF32Tensor { palette_f32: Vec<Vec<f32>>, // CLAM centroids in f32 entries: Vec<HhtlF32Entry>, slot_l: Option<Vec<SlotL>>, slot_l_scale: Option<f32>, svd_basis: Option<SvdBasis>, ... } impl HhtlF32Tensor { fn encode(role, rows, k) -> Self; // 1 B/row, argmax regime fn encode_with_leaf(role, rows, k, basis); // 9 B/row, index regime fn reconstruct_row(idx, n_cols) -> Vec<f32>; fn reconstruct_rows(n_cols) -> Vec<Vec<f32>>; } Pipeline: row → CLAM furthest-point → twig idx (1 byte) residual → SvdBasis::project → SlotL (8 × i8) decode: palette_f32[twig] + SvdBasis::reconstruct(slot_l * scale) Per-tensor footprint for [n_rows, n_cols]: palette BF16: 256 × n_cols × 2 SVD basis: 8 × n_cols × 2 entries: n_rows × 1 slot_l: n_rows × 8 (if index regime) Tests (5 new, all passing): encode_without_leaf_picks_real_rows_as_centroids reconstruct_without_leaf_returns_nearest_centroid encode_with_leaf_beats_without_leaf_on_real_rows ← ρ ≥ 0.95 on low-rank entry_byte_size_is_one storage_accounting_is_additive Example: universal_hhtl_f32_encode.rs — same gates as #183 universal encoder, but uses HhtlF32Tensor. Running on Qwen3-TTS-0.6B in background. ## Path B — cascade_attention_probe (codec-space inference) New example: cascade_attention_probe.rs. Measures argmax agreement between: Raw: argmax_i q · K[i]^T (f32 dot) Codec: argmax_i FisherZTable[pal_idx(q), pal_idx(K[i])] on 512 perturbed queries against a real attention K matrix (talker layer 0 self_attn.k_proj, shape [1024, 2048]). Pass criteria (subjective): ≥ 90% top-1 agreement → Path B viable for pipeline-wide swap ≥ 70% partial → Path B needs Q-side escalation layer < 70% fail → not competitive with f32 GEMM Both runs launched; results will be posted as PR comments when they complete. ## Session PR stack #180 merged SlotL foundation #181 merged HhtlDTensor × SlotL #182 merged SharedPaletteGroup × SlotL #183 merged universal_hhtld_encode (Base17 — reconstruction failure documented) #184 this PR HhtlF32Tensor codec + Path A/B examples https://claude.ai/code/session_01NYGrxVopyszZYgLBxe4hgj

Per codex P1 comment on #184: HhtlF32Entry.twig is u8, so valid centroid IDs are 0..=255. Before this fix, encode() accepted any k and assign_nearest_f32 silently wrapped ci as u8 — passing k=300 (say) would assign centroid-300 as twig-44 and reconstruct the wrong row. This was actively dangerous because the next-session plan (PR #184 thread) explicitly proposed k=1024 or 2048 centroids as the quality fallback. Fix: - New `pub const MAX_PALETTE_K: usize = 256` with clear docstring - Both `encode` and `encode_with_leaf` now assert: k > 0 k <= MAX_PALETTE_K with explicit panic messages naming the u8 twig limit Larger palettes need a codec with a wider twig-index (u16 would lift the cap to 65536, but changes the wire format). That's a separate PR if/when the quality probe shows k=512+ earns its keep. Tests (4 new, all pass + 5 existing): encode_rejects_zero_k (#[should_panic = "k > 0"]) encode_rejects_k_above_256 (#[should_panic = "u8 twig limit"]) encode_with_leaf_rejects_k_above_256 (same) encode_accepts_k_at_max_palette (k=256 must still succeed) Refs: - PR #184 codex P1 comment ("Reject palette sizes that exceed 255 centroids") - Follow-up to merged PRs #180/#181/#182/#183/#184 https://claude.ai/code/session_01NYGrxVopyszZYgLBxe4hgj

Session-end artefact for future déjà-vu. Catalogues every compression approach tried in PRs #176-#185 and the lesson each one produced. No approach is thrown away — each failed experiment carries information about where the real boundary is. ## Structure ### Core invariants (6) I1. Two regimes, opposite needs (argmax vs index) I2. Near-orthogonality of weight rows in high dim I3. Direction vs amplitude cannot be merged into one scalar I4. Wire-format type widths are hard caps — assert at encode time I5. 'u8 can span u16/u64 effective' requires the right decoder I6. The ticket-for-curve model (SpiralAddress + shared curve) ### Approaches tried (7) A1. HhtlDTensor — Base17 + Slot D + Slot V (correct for cascade, wrong for f32 GEMM) A2. Progressive residual RVQ with k-ladder (works argmax, fails index) A3. Hierarchical CLAM 256x256 (REFUTED — cos 0.0046 on vocab) A4. Passthrough BF16 n_rows > 8192 (SHIPS for correctness, net loss for ratio) A5. SlotL 8 x i8 on SVD basis (correct algorithm, misapplied to Base17 centroid) A6. HhtlF32Tensor f32 palette + SlotL (right direction, 10x better, still short) A7. cascade_attention_probe Base17 palette (3.71% argmax agreement — palette doesn't preserve inner products) ### Abstractions that ARE the right primitive (3) R1. highheelbgz::rehydrate::SpiralEncoding (exists, untested on real Qwen3) R2. Per-role stride in NeuronPrint (q/k=3, v=5, gate=8, up=2, down=4) R3. HHTL cascade inference (hhtl_cache RouteAction) ### Open probes (4) P1. SpiralEncoding on real Qwen3 weights — claim rho >= 0.95 unproven P2. Shared anchors + i8 position per row — depends on P1 P3. Palette preserves inner-product neighbourhoods — A7 refuted for Base17 P4. Log-radial CLAM with magnitude split — hypothesised > linear CLAM ### Déjà-vu table Lists 7 'if you're tempted to...' instincts with the PR that already refuted them. Exists so future sessions hit the lesson before writing the code. ### Structural checklist (5 questions) Before shipping any new codec: 1. What regime does this tensor belong to? (I1) 2. Does the codec encode direction AND amplitude separately? (I3) 3. Is the palette substrate inner-product-preserving? (I2, A7) 4. Does the decoder evaluate the curve, or tile anchors? (I5) 5. Are wire-format widths asserted at encode time? (I4) ## Why this doc matters Every failed approach in this session taught something the next session would otherwise re-learn the hard way. HCLAM (#177->#178) already has its lesson buried in a passthrough commit. The Base17 reconstruction failure (#183) is buried in a PR comment. The #184 Path A/B duality (they aren't independent) is only visible if you read the probe results. This doc surfaces all of it as a single index, structured for mutation: each approach has 'mutation hooks' naming how it could evolve into something that works, rather than being discarded. ## Next step blocked by token budget The SpiralEncoding-on-real-Qwen3 probe (P1) is the obvious next experiment and would have landed in this PR. Deferred to a fresh session with budget. The doc leaves the probe fully specified so re-entering cold loses no context. https://claude.ai/code/session_01NYGrxVopyszZYgLBxe4hgj

Codifies 7 anti-patterns (AP1-AP7) learned from PRs #176-#188 into an agent card that fires flags when the session repeats them: AP1: "225/225 feels like success" without gate 2 (#178) AP2: Projecting quality from docs instead of measuring (#177) AP3: Building new codec before benching existing ones (#184) AP4: Centroid-residual framing on near-orthogonal data (#177/#183) AP5: Python in the inference hot path AP6: Chained score multiplication without chain-collapse check (P5) AP7: Modifying ndarray without explicit permission (#176) Invoked by adk-coordinator when pattern repetition is suspected, or by human directly. Output: list of fired flags, max 7 lines. Also audited all 29 agent cards across both repos: - All pin model: opus or model: sonnet (no hardcoded versions) - opus → Opus 4.7 automatically, sonnet → Sonnet 4.6 - 3 ndarray agents on sonnet (l3-strategist, migration-tracker, product-engineer) — intentional for speed-over-depth roles - adk-coordinator missing Bash tool (by design — delegates) - sentinel-qa missing Edit/Write (by design — audit-only) No agent changes needed for Opus 4.7 compatibility — model: opus resolves correctly. https://claude.ai/code/session_01NYGrxVopyszZYgLBxe4hgj

claude added 2 commits April 15, 2026 06:31

chatgpt-codex-connector Bot reviewed Apr 15, 2026

View reviewed changes

AdaWorldAPI merged commit 4e0a8a6 into main Apr 15, 2026

AdaWorldAPI mentioned this pull request Apr 15, 2026

feat(bgz-tensor): HhtlF32Tensor codec + Path A encoder + Path B argmax-parity probe #184

Merged

AdaWorldAPI mentioned this pull request Apr 15, 2026

docs: codec invariants + experiment catalogue (session-end déjà-vu) #186

Merged

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

feat(examples): universal_hhtld_encode — model-generic encoder with SlotL dispatch#183

feat(examples): universal_hhtld_encode — model-generic encoder with SlotL dispatch#183
AdaWorldAPI merged 3 commits into
mainfrom
claude/universal-hhtld-encode

AdaWorldAPI commented Apr 15, 2026

Uh oh!

AdaWorldAPI commented Apr 15, 2026

Uh oh!

chatgpt-codex-connector Bot left a comment

Uh oh!

chatgpt-codex-connector Bot Apr 15, 2026

Uh oh!

chatgpt-codex-connector Bot Apr 15, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

		let total_output = total_entries_bytes + total_slot_l_bytes + total_svd_basis_bytes
		+ total_palette_bytes + passthrough_bytes;

		if tensor_rows_f32.is_empty() {
		return build_group_with_fisher_z(key, tensor_names, tensor_rows_f32, k);

Conversation

AdaWorldAPI commented Apr 15, 2026

Summary

What it does

Gates reported

Usage

Design notes

Stack

Test plan

Follow-ups if this PR passes

Follow-ups if GATES FAIL

Uh oh!

AdaWorldAPI commented Apr 15, 2026

Full run results: ratio passed, reconstruction gates failed

Root cause

What this DOES confirm

Path A — f32 centroid palette (reconstruction-grade)

Path B — inference in codec space (what HhtlDTensor is for)

Recommendation

Uh oh!

chatgpt-codex-connector Bot left a comment

Choose a reason for hiding this comment

💡 Codex Review

Uh oh!

chatgpt-codex-connector Bot Apr 15, 2026

Choose a reason for hiding this comment

Uh oh!

chatgpt-codex-connector Bot Apr 15, 2026

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants