gc4b (v11 beta) q4f16: new shards + wasm + version.json bump v10->v11 (cache-bust)

v10 (cross-attribution v2): gc3 q4f16 clean weight-swap — new params + tensor-cache + version v9->v10 (wasm/config unchanged)

v9 (persona-sink fix): gc2 q4f16 clean weight-swap — new params + tensor-cache + version v8->v9 (wasm/config/tokenizer unchanged)

gc1 -> v8 (grounding-clean): weight-swap + version bump v7->v8; worn P +5pts on gold, removal 38/38, drift Q4 |Δ|<=0.04

v7 cross-attribution build (54->85%); weights+config+version, wasm unchanged

tkaug (v0.6, removal-capable): q4f16 bundle, version 0.6.0-tkaug

release v6: rc-cpc2-pad (worn 0.67) — bump version+modelId to trigger update + bust WebLLM cache

rc-cpc2-pad (0.8B, worn 0.67 @ max-new 2048; padding+bronze recipe) — overwrite 2026-07-05

v5: MIN short-prompt build (q4f16) + clean README

version.json -> v4 (temp 0.1)

config: temperature 0.1 (near-greedy) — temp 0 hit a WebLLM divide-by-temp edge case

version.json -> v3 (greedy-config fix)

config: temperature 0 (greedy) — deterministic extractor parity

Drop throwaway/versioned wasm names

version.json: modelLib -> stable name

Stable model-lib name (Beholder-q4f16-webgpu.wasm)

Bump version.json -> v2

v2 WebGPU model lib (right-sized)

v2 config: 8K context, prefill 512, batch 1 (~1.1GB VRAM)

version.json: add ggufUrl for the update dialog

Production framing + version note (drop early-test caveat)

Add version.json (v1) for the extension auto-update flow

Browser/WebLLM-specific card (split from GGUF)

Add throwaway Qwen3.5 q4f16 WebLLM bundle (weights + config + tokenizer + WebGPU .wasm)

Add model card + PolyForm-NC license metadata

Delete README.md

Update README.md

Update README.md

Update README.md

Create README.md

initial commit