| { | |
| "id": "flat_llm", | |
| "name": "FLAT-LLM: Activation-Aware Structured Dimension & Neuron Pruning via Eigen-", | |
| "file": "Codes/framework/compression_engine/flat_llm.py", | |
| "type": "Compression Engine", | |
| "category": "Structured Pruning & Channel Slicing", | |
| "paradigm": "pruning", | |
| "icon": "\u2702\ufe0f", | |
| "description": "FLAT-LLM: Activation-Aware Structured Dimension & Neuron Pruning via Eigen-Decomposition. Registered in COMPRESSOR_REGISTRY as 'flat_llm' and 'flat_llm_compressor'. Production hardened for multi-architecture compatibility (DistilBERT, RoBERTa, GPT-2, Qwen3.5) with reconstruction validation, Conv1D support, and safe fallback.", | |
| "classes": [ | |
| "FLATLLMCompressor" | |
| ], | |
| "methods_count": 8, | |
| "math_latex": "\\min_{W_r} \\text{Tr}((W - W_r) R_{XX} (W - W_r)^\\top)", | |
| "components": [ | |
| { | |
| "id": "orthogonal_rot", | |
| "name": "Orthogonal Transformation Matrix (Q)", | |
| "role": "Energy Concentration", | |
| "shape_transform": "[d_{in} \u00d7 d_{in}] Orthogonal", | |
| "color": "#38bdf8", | |
| "math_latex": "X' = X Q, \\quad Q^\\top Q = I" | |
| }, | |
| { | |
| "id": "sliced_linear", | |
| "name": "Channel-Sliced Dense Linear Operator", | |
| "role": "45% Channel Elimination", | |
| "shape_transform": "[0.55 \\cdot d_{out} \u00d7 0.55 \\cdot d_{in}]", | |
| "color": "#0ea5e9", | |
| "math_latex": "W_{sliced} = (Q_2^\\top W Q_1)_{[:d'_1, :d'_2]}" | |
| }, | |
| { | |
| "id": "bias_correct", | |
| "name": "Mean Bias Correction Vector", | |
| "role": "Representation Recalibration", | |
| "shape_transform": "[0.55 \\cdot d_{out}]", | |
| "color": "#0284c7", | |
| "math_latex": "\\mu_{err} = \\mathbb{E}[W X] - \\mathbb{E}[W_{sliced} X']" | |
| } | |
| ], | |
| "model_projections": { | |
| "distilbert": { | |
| "hidden_size": 768, | |
| "intermediate_size": 3072, | |
| "baseline_layer_params": 6488064, | |
| "compressed_layer_params": 1946419, | |
| "savings_ratio": 3.33 | |
| }, | |
| "roberta": { | |
| "hidden_size": 768, | |
| "intermediate_size": 3072, | |
| "baseline_layer_params": 6488064, | |
| "compressed_layer_params": 1946419, | |
| "savings_ratio": 3.33 | |
| }, | |
| "gpt2": { | |
| "hidden_size": 768, | |
| "intermediate_size": 3072, | |
| "baseline_layer_params": 6488064, | |
| "compressed_layer_params": 1946419, | |
| "savings_ratio": 3.33 | |
| }, | |
| "qwen": { | |
| "hidden_size": 896, | |
| "intermediate_size": 4864, | |
| "baseline_layer_params": 11124736, | |
| "compressed_layer_params": 3337420, | |
| "savings_ratio": 3.33 | |
| } | |
| } | |
| } |
Xet Storage Details
- Size:
- 2.49 kB
- Xet hash:
- 054029f1262234076268de1ba412b090718b39f68b5efa374af31efbc8be6f33
·
Xet efficiently stores files, intelligently splitting them into unique chunks and accelerating uploads and downloads. More info.