Commit History
-
fix: render AdaRound FP8 loss as $$\min \|Wx - \tilde{W}(V)x\|_2^2$$ and add blue flower limitation disclaimer (Brunonia/Dampiera validation)
thenukegun10x6f356ab
-
docs: add FP8 hardware warning to Server/Desktop GPU/CPU
Verified signatureFP8 E4M3 not commonly supported - requires Ada/Hopper RTX 40xx/L4/H100 CUDA12+ PyTorch2.1+. Add warning to B. Server/Desktop table: check hardware or if unsure use BF16 (identical accuracy 89.31%/99.42% vs 104MB). Requested edit.05f0f04
-
Re-bench 2/3/4-view sequential BF16 2026-08-30: Dense 4View 98.31% / MoE BEST 99.42% (bench_multiview_finals.json)
Verified signature237bb08 -
Update Model Card to match exact FP8-AdaRound and standard INT8 repo filenames
Verified signature755ca51 -
Add direct download links, accurate AdaRound FP8 vs INT8 quantization specs, and clean filenames
Verified signature9031464 -
Update Model Card with verified Dense 4-View multi-angle benchmark results
Verified signaturef14b921 -
Add CLI guide and quickstart instructions to Model Card
Verified signaturef1ffdeb -
Update Multi-View benchmark table to mark Dense 4-View as Training in Progress
Verified signature61f63c0 -
Clarify 4-View dedicated weights and add Dense 4-View Distilled specifications
Verified signature4edfcae
-
Update Model Card with Dense Multi-View benchmarks and complete artifact catalog
Verified signaturecd43d04 -
Add Dense Mobile & Edge Suite with ONNX and LiteRT models
Verified signature00ec82c -
Add multi-photo (4 angles required) documentation and code examples
Verified signaturee329a6e -
Update Model Card with PlantDetect naming and AdaRound-FP8 methodology
Verified signature1de07f5 -
Initial commit: Model Card, Species Index (999 spp), and Model Architecture
Verified signature3217f9a