Confirm all checkpoint artifacts verified and publish usage notes
Browse files- README.md +11 -1
- UPLOAD_VERIFICATION.json +12 -0
README.md
CHANGED
|
@@ -10,7 +10,7 @@ tags:
|
|
| 10 |
|
| 11 |
# VPP2 compact recovery — update 1,000
|
| 12 |
|
| 13 |
-
|
| 14 |
|
| 15 |
This is the **physically pruned policy checkpoint selected for the full-suite evaluation campaign**. The uploaded inference files match the SHA-256 fingerprints in all five suite protocols (Precision, Memory, Generalization, Long Horizon and Open). This establishes artifact identity, not completion of every evaluation episode.
|
| 16 |
|
|
@@ -32,6 +32,16 @@ At update 1,000, the fixed 140-window offline validation panel reported task-mac
|
|
| 32 |
|
| 33 |
The original `COMPLETE.json` has `simulation_completed: false`, reflecting its export-time status. It is preserved unedited; subsequent evaluation provenance is separate.
|
| 34 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 35 |
## Lineage and code
|
| 36 |
|
| 37 |
Mask learning update 1,500 selected the channels. The physically pruned video expert has **12,468,409,664 parameters**, compared with 16,394,878,784 before pruning (23.95% fewer video parameters). The frozen Action2B expert is separate from this count. Recovery checkpoint 55 already had 55 recovery updates with K/V supervision; the later no-K/V branch continued from it to update 1,000.
|
|
|
|
| 10 |
|
| 11 |
# VPP2 compact recovery — update 1,000
|
| 12 |
|
| 13 |
+
Complete: listed model artifacts verified against local receipts and remote checksums.
|
| 14 |
|
| 15 |
This is the **physically pruned policy checkpoint selected for the full-suite evaluation campaign**. The uploaded inference files match the SHA-256 fingerprints in all five suite protocols (Precision, Memory, Generalization, Long Horizon and Open). This establishes artifact identity, not completion of every evaluation episode.
|
| 16 |
|
|
|
|
| 32 |
|
| 33 |
The original `COMPLETE.json` has `simulation_completed: false`, reflecting its export-time status. It is preserved unedited; subsequent evaluation provenance is separate.
|
| 34 |
|
| 35 |
+
## Download
|
| 36 |
+
|
| 37 |
+
Authenticate with an account that has access to this private repository, then run:
|
| 38 |
+
|
| 39 |
+
```bash
|
| 40 |
+
hf download GnanaPraveen/vpp2-recovery-step-1000 --local-dir /data3/praveen/models/vpp2-recovery-step-1000
|
| 41 |
+
```
|
| 42 |
+
|
| 43 |
+
Use the code and setup link below to load these files. Keep the integrity receipts with the downloaded weights.
|
| 44 |
+
|
| 45 |
## Lineage and code
|
| 46 |
|
| 47 |
Mask learning update 1,500 selected the channels. The physically pruned video expert has **12,468,409,664 parameters**, compared with 16,394,878,784 before pruning (23.95% fewer video parameters). The frozen Action2B expert is separate from this count. Recovery checkpoint 55 already had 55 recovery updates with K/V supervision; the later no-K/V branch continued from it to update 1,000.
|
UPLOAD_VERIFICATION.json
ADDED
|
@@ -0,0 +1,12 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"artifact_revision": "bc72b758356042311edf1592171d48c33e5eeb90",
|
| 3 |
+
"bytes": 29002345549,
|
| 4 |
+
"files": 16,
|
| 5 |
+
"inventory_sha256": "6b27ac39a5c9e96e3ea45be5da6d32006e393a75fd8b3eb921a714bc4c44e1b3",
|
| 6 |
+
"optimizer_resume_shards_uploaded": false,
|
| 7 |
+
"original_files_modified": false,
|
| 8 |
+
"private": true,
|
| 9 |
+
"repo_id": "GnanaPraveen/vpp2-recovery-step-1000",
|
| 10 |
+
"verification": "Source SHA-256 verified against recorded receipts; remote LFS SHA-256 or Git blob matched for every artifact; source size/mtime unchanged.",
|
| 11 |
+
"verified_at": "2026-11-04T20:00:51.244557+00:00"
|
| 12 |
+
}
|