Update README.md

README.md

@@ -1,34 +1,3 @@
-# LFM-2.5-1.2B-Compact-Reservoir
-
-**Tags:** triton, svd-compression, spiking-neuron, liquid-foundation-model, low-rank-adaptation, efficient-inference  
-**Library:** transformers  
-**Pipeline:** text-generation
-
-Replaces MLP layers with custom `CompactReservoirFFN`: SVD compression (rank_ratio=0.25), Triton kernels, spiking neuron activation.
-
-## ✨ Key Features
-- **SVD Compression**: MLP weights reduced 4x, preserves performance.  
-- **Triton Kernels**: `fast_einsum_gating` for dynamic merging, low memory/latency.  
-- **Spiking Neurons**: Membrane potential  
-
-  \[
-  V_m[t] = \tau V_m[t-1] + (1-\tau)\text{Mean}(Y)
-  \]  
-
-  spike if \(V_m > 0.5\).  
-- **Context Gating**:  
-
-  \[
-  G = \text{Softmax}(\text{Gate}(X + \text{ContextBias})), \quad W_{\text{merged}} = \sum G_k \cdot W_{\text{svd}_k}
-  \]
-
-## 📊 Benchmarks
-20 prompts: reasoning / coding / writing
-
-| Model                  | Latency | Memory (MLP) |
-|------------------------|---------|--------------|
-| Original LFM-2.5-1.2B  | 1.00x   | 100%         |
-| Compact-Reservoir       | 0.85x   | 75%          |
 
 ## 🚀 Quick Start