| { | |
| "id": "astro_quantizer", | |
| "name": "AstroQuantizer: Dynamic Uniform Range Quantization Engine with Straight-Thr", | |
| "file": "Codes/framework/compression_engine/astro_quantizer.py", | |
| "type": "Compression Engine", | |
| "category": "Low-Rank SVD & Factorization", | |
| "paradigm": "svd", | |
| "icon": "\u26a1", | |
| "description": "AstroQuantizer: Dynamic Uniform Range Quantization Engine with Straight-Through Estimator (STE) and Fisher-Adaptive Mixed-Precision Bit Allocation. Registered as 'astro' and 'astro_quantizer'. Scientific Novelty: - Mixed-precision dynamic bit allocation driven by Empirical Fisher Information / Output Activation Energy. - Optimal integer budget allocation: Channels with higher Fisher sensitivity r", | |
| "classes": [ | |
| "STEFakeQuantize", | |
| "AstroLinear", | |
| "AstroQuantizer" | |
| ], | |
| "methods_count": 15, | |
| "math_latex": "W \\approx U_r \\Sigma_r V_r^\\top = A B^\\top, \\quad r \\ll \\min(d_{in}, d_{out})", | |
| "components": [ | |
| { | |
| "id": "v_factor", | |
| "name": "Factor V (Low-Rank Subspace Projection)", | |
| "role": "Projection to Rank r", | |
| "shape_transform": "[r \u00d7 d_in]", | |
| "color": "#3acbb8", | |
| "math_latex": "z_t = x_t V_r^\\top, \\quad V_r \\in \\mathbb{R}^{r \\times d_{in}}" | |
| }, | |
| { | |
| "id": "u_factor", | |
| "name": "Factor U (Full-Rank Expansion Projection)", | |
| "role": "Expansion to Output Space", | |
| "shape_transform": "[d_out \u00d7 r]", | |
| "color": "#2dd4bf", | |
| "math_latex": "y_t = z_t (\\Sigma_r U_r^\\top) + b, \\quad U_r \\in \\mathbb{R}^{d_{out} \\times r}" | |
| }, | |
| { | |
| "id": "adapter", | |
| "name": "Residual LoRA / Norm Calibration", | |
| "role": "Fidelity Compensation", | |
| "shape_transform": "[d_out \u00d7 d_in] Residual", | |
| "color": "#5eead4", | |
| "math_latex": "\\Delta y_t = x_t (B A)^\\top \\cdot \\frac{\\alpha}{r}, \\quad A \\in \\mathbb{R}^{r \\times d_{in}}, \\; B \\in \\mathbb{R}^{d_{out} \\times r}" | |
| } | |
| ], | |
| "model_projections": { | |
| "distilbert": { | |
| "hidden_size": 768, | |
| "intermediate_size": 3072, | |
| "baseline_layer_params": 6488064, | |
| "compressed_layer_params": 786432, | |
| "savings_ratio": 8.25 | |
| }, | |
| "roberta": { | |
| "hidden_size": 768, | |
| "intermediate_size": 3072, | |
| "baseline_layer_params": 6488064, | |
| "compressed_layer_params": 786432, | |
| "savings_ratio": 8.25 | |
| }, | |
| "gpt2": { | |
| "hidden_size": 768, | |
| "intermediate_size": 3072, | |
| "baseline_layer_params": 6488064, | |
| "compressed_layer_params": 786432, | |
| "savings_ratio": 8.25 | |
| }, | |
| "qwen": { | |
| "hidden_size": 896, | |
| "intermediate_size": 4864, | |
| "baseline_layer_params": 11124736, | |
| "compressed_layer_params": 1081344, | |
| "savings_ratio": 10.29 | |
| } | |
| } | |
| } |
Xet Storage Details
- Size:
- 2.74 kB
- Xet hash:
- 1cdae4b568daf4849021d6aa459c6b74d85bbb32c879823e43473c2b300d35b0
·
Xet efficiently stores files, intelligently splitting them into unique chunks and accelerating uploads and downloads. More info.