[WIP] perf(policies): Make ACT policy compatible with torch.compile

benchmarks/policies_compilation/baseline_act_report.md

@@ -0,0 +1,97 @@
+# Torch.compile Benchmark Report: ACT
+
+## Environment
+
+- **Policy**: act
+- **Device**: cuda
+- **PyTorch**: 2.7.1+cu126
+- **Dataset**: AdilZtn/grab_red_cube_test_25
+- **Batch Size**: 8
+- **Benchmark Parameters**: 100 inference runs, 50 training runs
+
+## 🔧 Compilation Results
+
+- **Status**: ✅ SUCCESS
+
+## 🎯 Correctness Results
+
+- **Status**: ❌ FAILED
+- **Inference**: FAILED
+- **Training**: FAILED
+
+### Detailed Differences
+
+- **Max Action Difference**: 5.57e-02 (threshold: 1.00e-05)
+- **Loss Difference**: 8.93e-05 (threshold: 1.00e-05)
+
+### ⚠️ Correctness Analysis
+
+- **Action diff magnitude**: 5.57e-02 (SEVERE)
+- **Loss diff magnitude**: 8.93e-05 (MINOR)
+- **Likely causes**: Graph breaks, dynamic shapes, numerical precision issues
+
+## ⚡ Performance Results
+
+### Inference Performance
+
+- **Original**: 21.75 ms/iter
+- **Compiled**: 21.46 ms/iter
+- **🚀 Speedup**: 1.01x (⚠️ INSUFFICIENT)
+
+### Training Performance
+
+- **Original**: 68.59 ms/iter
+- **Compiled**: 61.15 ms/iter
+- **🚀 Speedup**: 1.12x
+
+### Consistency Metrics
+
+- **Average Loss Difference**: 4.87e-03
+- **Average Grad Norm Difference**: 1.60e+00
+
+## 📋 Success Criteria Analysis
+
+- **✅ Compilation**: PASSED
+- **✅ Correctness**: FAILED
+- **✅ Performance**: FAILED
+- **✅ Benchmarking**: PASSED
+
+## 🎯 Overall Result
+
+❌ NEEDS WORK: torch.compile not yet functional
+
+## 🛠️ Next Steps
+
+1. **Debug numerical differences** - Check for precision issues
+2. **Verify tensor operations** - Ensure deterministic behavior
+3. **Test with smaller tolerance** - May be acceptable for some use cases
+
+## 🔍 Raw Data
+
+```json
+{
+  "success": false,
+  "policy": "act",
+  "device": "cuda",
+  "pytorch_version": "2.7.1+cu126",
+  "compilation_successful": true,
+  "compilation_error": null,
+  "correctness": {
+    "inference_correct": false,
+    "training_correct": false,
+    "action_diff": 0.05568695068359375,
+    "loss_diff": 8.934736251831055e-5
+  },
+  "correctness_passed": false,
+  "inference_benchmarked": true,
+  "training_benchmarked": true,
+  "time_original_inference": 21.745667571667582,
+  "time_compiled_inference": 21.46407115040347,
+  "speedup_inference": 1.0131194319703334,
+  "time_original_training": 68.5850445041433,
+  "time_compiled_training": 61.15469123702496,
+  "speedup_training": 1.1215009530228772,
+  "loss_consistency": 0.004866420030593872,
+  "grad_norm_consistency": 1.599840692281723
+}
+```