Verify portable CPU training path and pin public training assets

training/trackformer_1_2/EXECUTION_CHECKS.json

@@ -0,0 +1,35 @@
+{
+  "status": "passed",
+  "device": "cpu",
+  "optimizer_updates": 0,
+  "full_training_rerun": false,
+  "samples": [
+    {
+      "native": false,
+      "finite_backward": true,
+      "attention_gradient_nonzero": true,
+      "field_pressure_exact": true,
+      "future_input_rejected": true
+    },
+    {
+      "native": true,
+      "finite_backward": true,
+      "attention_gradient_nonzero": true,
+      "field_pressure_exact": true,
+      "future_input_rejected": true
+    }
+  ],
+  "sampler_all_train_rows": true,
+  "counts": {
+    "train": 13949,
+    "validation": 1041,
+    "test": 1195
+  },
+  "native_counts": {
+    "train": 800,
+    "validation": 100,
+    "test": 100
+  },
+  "preflight": "passed",
+  "new_full_training_run": false
+}

training/trackformer_1_2/PUBLICATION.md

@@ -0,0 +1,19 @@
+# Immutable publication and execution checks
+
+- Source publication on GitHub: https://github.com/yu314-coder/typhoon-predict/commit/ac05e1974ac90ad7819bf0483da8e8637d880525
+- Source and all nine assets on Hugging Face: https://huggingface.co/euler314/typhoon-predict/commit/2d1464e9a310e6a6e8840ef1c23017fdededd57b
+- All asset hashes remain fixed in `assets.json`.
+
+Pin the binary download to the completed publication:
+
+```bash
+python training/trackformer_1_2/download_assets.py --revision 2d1464e9a310e6a6e8840ef1c23017fdededd57b
+```
+
+The original trainer's CPU preflight passed, including all six input-file hashes, exact split counts, normalization and original warm-start loading (21,452,595 parameters). The portable CPU check passed on one coarse and one native TRAIN example: finite forward/loss/backward, nonzero attention gradient, pressure exactly read from the predicted field, rejection of an added future input, and full TRAIN sampler coverage. Zero optimizer updates were performed. See `EXECUTION_CHECKS.json`. This is an execution check, not a new end-to-end training reproduction or an accuracy claim.
+
+```bash
+python training/trackformer_1_2/test_training_cpu.py
+```
+
+This test also reconstructs the loader's local basin mmap; use a dedicated writable data copy, never another trainer's data directory. Original GPU-specific tests are archived without rewriting their historical relative-path assumptions in `source/v173/test_model.py`; they were not rerun on GPU for this publication.

training/trackformer_1_2/SOURCE_MANIFEST.json

@@ -3,40 +3,45 @@
   "unchanged_original_sources": true,
   "model_weights_modified": false,
   "files": [
+    {
+      "path": "training/trackformer_1_2/EXECUTION_CHECKS.json",
+      "sha256": "baa5800cdca8cbcffdfd5380382d76722613f76169cadf152a4d32553da9c09f",
+      "bytes": 738
+    },
     {
       "path": "training/trackformer_1_2/LICENSE",
       "sha256": "090a97ebbccef9de77cb3898c473f5c9ada7e140777957f1195fc6ab89b9a313",
       "bytes": 1083
     },
     {
-      "path": "training/trackformer_1_2/requirements.txt",
-      "sha256": "37e15a665eff6c8ee94c47ac2da0f7c7e2d699b7e21623b0289da4cf5d3995d0",
-      "bytes": 45
+      "path": "training/trackformer_1_2/PUBLICATION.md",
+      "sha256": "1bcaa51a1999422925db2a100927985b927b454a2ff4250e318f7049ceec81c8",
+      "bytes": 1505
     },
     {
-      "path": "training/trackformer_1_2/config.json",
-      "sha256": "813f5d05a3dec5ec160b4e91eb80970fb11d88416fc7709ffc9d147fc880316b",
-      "bytes": 820
+      "path": "training/trackformer_1_2/README.md",
+      "sha256": "342c43d94710355101ddc5bcad44923f47707f473eb803398f44fe2d77e8ace9",
+      "bytes": 6560
     },
     {
       "path": "training/trackformer_1_2/THIRD_PARTY_NOTICES.md",
       "sha256": "f764dd203eb124fa6192edd3b7b49a7a546cd5d69d3a8a250589a14ce43e2115",
       "bytes": 1697
     },
+    {
+      "path": "training/trackformer_1_2/assets.json",
+      "sha256": "a59148e9ba2617599614784b47789c3654df1a7809ee039e78bcd00fe982b211",
+      "bytes": 1993
+    },
     {
       "path": "training/trackformer_1_2/checkpoint-history.json",
       "sha256": "8b9593a7ed58c1bd28dd27c8049c89a0c5c5644a30bfe5203a4b54c445ff2fb4",
       "bytes": 459
     },
     {
-      "path": "training/trackformer_1_2/README.md",
-      "sha256": "342c43d94710355101ddc5bcad44923f47707f473eb803398f44fe2d77e8ace9",
-      "bytes": 6560
-    },
-    {
-      "path": "training/trackformer_1_2/train.py",
-      "sha256": "4502a87bfafb11667015070c6fad37b723b4a53723789febda075fba1c7770e8",
-      "bytes": 1436
+      "path": "training/trackformer_1_2/config.json",
+      "sha256": "813f5d05a3dec5ec160b4e91eb80970fb11d88416fc7709ffc9d147fc880316b",
+      "bytes": 820
     },
     {
       "path": "training/trackformer_1_2/download_assets.py",
@@ -44,39 +49,29 @@
       "bytes": 3049
     },
     {
-      "path": "training/trackformer_1_2/assets.json",
-      "sha256": "a59148e9ba2617599614784b47789c3654df1a7809ee039e78bcd00fe982b211",
-      "bytes": 1993
-    },
-    {
-      "path": "training/trackformer_1_2/source/v173/test_model.py",
-      "sha256": "493e0db88a098c89bb4db5e6c577fa24b4b7bcb83ce547dceac35237dbce7206",
-      "bytes": 5579
-    },
-    {
-      "path": "training/trackformer_1_2/source/v173/v165_base.py",
-      "sha256": "5c9473993a19d0fed00055494b5c1f50c1006aaf46551ff50145f67048c3374a",
-      "bytes": 39040
+      "path": "training/trackformer_1_2/requirements.txt",
+      "sha256": "37e15a665eff6c8ee94c47ac2da0f7c7e2d699b7e21623b0289da4cf5d3995d0",
+      "bytes": 45
     },
     {
-      "path": "training/trackformer_1_2/source/v173/v173_pressure_train.py",
-      "sha256": "f40691cec3e787f4ac3c189f647efb0953c1eaa7e03fc86edaf4fee06990b2be",
-      "bytes": 11200
+      "path": "training/trackformer_1_2/source/data_preparation/requirements.txt",
+      "sha256": "54918b8f29330299e18b8924e9cd93d05807e7f8cd99b2efd4805e9be94003c1",
+      "bytes": 55
     },
     {
-      "path": "training/trackformer_1_2/source/v173/model.py",
-      "sha256": "5dab06c20224cf48c51302122455ee7e16948f977bf4b760fd8aff5d443ab0bc",
-      "bytes": 3428
+      "path": "training/trackformer_1_2/source/data_preparation/v164_dataset_contract.md",
+      "sha256": "cfd838651e632aba0d2439ba202a59895bd4df22137185974df1d97f0e2525d4",
+      "bytes": 4664
     },
     {
-      "path": "training/trackformer_1_2/source/v173/DESIGN.md",
-      "sha256": "163ae39ff572f23c519d65e27ee6a8db23b47df9461670709229ccf95130049a",
-      "bytes": 3278
+      "path": "training/trackformer_1_2/source/data_preparation/v164_download_pressure_geography.py",
+      "sha256": "b8602cd2fba131a287dc9e25781c8f037d3325fec8a23203576916c2f7d2189c",
+      "bytes": 9256
     },
     {
-      "path": "training/trackformer_1_2/source/v173/baseline_model.py",
-      "sha256": "6c05941fd93baaca04508877edc69f8f0394b70e0a64e734ef0d2c1e8832246a",
-      "bytes": 10040
+      "path": "training/trackformer_1_2/source/data_preparation/v164_global_regional_train.py",
+      "sha256": "c0808a1467695d2bc2ad8483aa4f4c6de1463628221024ac559f6204effe7978",
+      "bytes": 52581
     },
     {
       "path": "training/trackformer_1_2/source/data_preparation/v164_prepare_data.py",
@@ -84,9 +79,14 @@
       "bytes": 19312
     },
     {
-      "path": "training/trackformer_1_2/source/data_preparation/requirements.txt",
-      "sha256": "54918b8f29330299e18b8924e9cd93d05807e7f8cd99b2efd4805e9be94003c1",
-      "bytes": 55
+      "path": "training/trackformer_1_2/source/data_preparation/v164_prepare_mac.py",
+      "sha256": "7e7313f865369d42f67645196b4af0e1b2b84216303f7b2b61e577d3734df79a",
+      "bytes": 1309
+    },
+    {
+      "path": "training/trackformer_1_2/source/data_preparation/v164_reuse_impl.py",
+      "sha256": "2da0078d4a74faa93f8cf874e42707b01f286406f0ceaa4db1814aab0eceb7d1",
+      "bytes": 18414
     },
     {
       "path": "training/trackformer_1_2/source/data_preparation/v164_structure_research.md",
@@ -94,29 +94,44 @@
       "bytes": 17119
     },
     {
-      "path": "training/trackformer_1_2/source/data_preparation/v164_download_pressure_geography.py",
-      "sha256": "b8602cd2fba131a287dc9e25781c8f037d3325fec8a23203576916c2f7d2189c",
-      "bytes": 9256
+      "path": "training/trackformer_1_2/source/v173/DESIGN.md",
+      "sha256": "163ae39ff572f23c519d65e27ee6a8db23b47df9461670709229ccf95130049a",
+      "bytes": 3278
     },
     {
-      "path": "training/trackformer_1_2/source/data_preparation/v164_prepare_mac.py",
-      "sha256": "7e7313f865369d42f67645196b4af0e1b2b84216303f7b2b61e577d3734df79a",
-      "bytes": 1309
+      "path": "training/trackformer_1_2/source/v173/baseline_model.py",
+      "sha256": "6c05941fd93baaca04508877edc69f8f0394b70e0a64e734ef0d2c1e8832246a",
+      "bytes": 10040
     },
     {
-      "path": "training/trackformer_1_2/source/data_preparation/v164_dataset_contract.md",
-      "sha256": "cfd838651e632aba0d2439ba202a59895bd4df22137185974df1d97f0e2525d4",
-      "bytes": 4664
+      "path": "training/trackformer_1_2/source/v173/model.py",
+      "sha256": "5dab06c20224cf48c51302122455ee7e16948f977bf4b760fd8aff5d443ab0bc",
+      "bytes": 3428
     },
     {
-      "path": "training/trackformer_1_2/source/data_preparation/v164_global_regional_train.py",
-      "sha256": "c0808a1467695d2bc2ad8483aa4f4c6de1463628221024ac559f6204effe7978",
-      "bytes": 52581
+      "path": "training/trackformer_1_2/source/v173/test_model.py",
+      "sha256": "493e0db88a098c89bb4db5e6c577fa24b4b7bcb83ce547dceac35237dbce7206",
+      "bytes": 5579
     },
     {
-      "path": "training/trackformer_1_2/source/data_preparation/v164_reuse_impl.py",
-      "sha256": "2da0078d4a74faa93f8cf874e42707b01f286406f0ceaa4db1814aab0eceb7d1",
-      "bytes": 18414
+      "path": "training/trackformer_1_2/source/v173/v165_base.py",
+      "sha256": "5c9473993a19d0fed00055494b5c1f50c1006aaf46551ff50145f67048c3374a",
+      "bytes": 39040
+    },
+    {
+      "path": "training/trackformer_1_2/source/v173/v173_pressure_train.py",
+      "sha256": "f40691cec3e787f4ac3c189f647efb0953c1eaa7e03fc86edaf4fee06990b2be",
+      "bytes": 11200
+    },
+    {
+      "path": "training/trackformer_1_2/test_training_cpu.py",
+      "sha256": "35e9379c40f5928f72f8267dcbdbea892387e0261d74e476730ddcbdf143ba1b",
+      "bytes": 3279
+    },
+    {
+      "path": "training/trackformer_1_2/train.py",
+      "sha256": "4502a87bfafb11667015070c6fad37b723b4a53723789febda075fba1c7770e8",
+      "bytes": 1436
     }
   ],
   "assets": [

training/trackformer_1_2/test_training_cpu.py

@@ -0,0 +1,69 @@
+"""CPU-only execution check of real TRAIN inputs; zero optimizer updates."""
+import argparse
+import json
+from pathlib import Path
+import sys
+
+
+def main():
+    here = Path(__file__).resolve().parent
+    p = argparse.ArgumentParser(description=__doc__)
+    p.add_argument('--data-root', type=Path, default=here / 'data')
+    p.add_argument('--warm-start', type=Path, default=here / 'checkpoints/v168_epoch_003.pt')
+    a = p.parse_args()
+    sys.path.insert(0, str(here / 'source/v173'))
+    import torch
+    from model import CoreForecaster, objective, base
+    from v173_pressure_train import warm_start, batches
+    torch.set_num_threads(2)
+    torch.manual_seed(168)
+    base.EvolutionNet.block = lambda self, module, x: module(x)
+    m = base.load_reuse(a.data_root / 'manifest.json')
+    model = CoreForecaster(m).cpu()
+    warm_start(model, a.warm_start, m)
+    data = base.ReuseDataset(m, 'train')
+    samples = [next(i for i, r in enumerate(data.rows) if (r['patch'] >= 0) == native and r['issue_mask'][1])
+               for native in (False, True)]
+    results = []
+    for index in samples:
+        x, y = data[index]
+        x = {k: v[None] for k, v in x.items()}
+        model.train()
+        model.zero_grad(set_to_none=True)
+        state = model.initial(x)
+        state, prediction = model.step(state)
+        target = {k: v[0:1] for k, v in y.items()}
+        target.update(previous_pressure=x['issue_intensity'][:, 1],
+                      previous_pressure_mask=x['issue_mask'][:, 1],
+                      previous_center=x['center'],
+                      previous_storm_mask=torch.ones_like(target['storm_mask']))
+        loss, terms = objective(prediction, target, x, m)
+        assert bool(torch.isfinite(loss))
+        loss.backward()
+        assert all(bool(torch.isfinite(v.grad).all()) for v in model.parameters() if v.grad is not None)
+        assert model.context.output.weight.grad.abs().sum() > 0
+        grid = model.core_grid(prediction['center'][:, 0, None, None],
+                               prediction['center'][:, 1, None, None], state['origin'])
+        pressure = base.sample_field(prediction['core'] * model.std[:, :1] + model.mean[:, :1], grid)[:, 0, 0, 0]
+        assert torch.equal(pressure, prediction['pressure'])
+        try:
+            model.initial(dict(x, future_pressure=target['regional']))
+        except ValueError:
+            pass
+        else:
+            raise AssertionError('Future target accepted as input')
+        results.append(dict(native=bool(x['detail_available'].item()), finite_backward=True,
+                            attention_gradient_nonzero=True, field_pressure_exact=True,
+                            future_input_rejected=True))
+    indices = base.training_indices(data.rows, 1)
+    assert set(indices) == set(range(len(data)))
+    for parts, count in batches(indices, data.rows, 10, 4):
+        assert sum(map(len, parts)) == count
+        assert all(len({data.rows[i]['patch'] >= 0 for i in part}) == 1 for part in parts)
+    print(json.dumps(dict(status='passed', device='cpu', optimizer_updates=0,
+        full_training_rerun=False, samples=results, sampler_all_train_rows=True,
+        counts=m['counts'], native_counts=m['patch_counts']), indent=2))
+
+
+if __name__ == '__main__':
+    main()