wuschelschulz commited on
Commit
1267db1
·
verified ·
1 Parent(s): 55e998f

Publish completed overnight evaluation report

Browse files
README.md CHANGED
@@ -9,6 +9,8 @@ tags:
9
 
10
  Custom PyTorch `state_dict` checkpoint for the [GIPF Zero source](https://github.com/wusche1/gipf-zero). Load it with `training.model.load_model`.
11
 
 
 
12
  Source training checkpoint SHA-256: `aa533bb8922231a3e0dc53dd2ff947cab6899ef09a9245463c89ac99ea776ef0`.
13
 
14
  Served `champion.pt` artifact SHA-256: `d5a7e3bda2966a79d22cf34f131976907260e679683d0e2680d935312f41c2f6`.
 
9
 
10
  Custom PyTorch `state_dict` checkpoint for the [GIPF Zero source](https://github.com/wusche1/gipf-zero). Load it with `training.model.load_model`.
11
 
12
+ Published public opponent: **GIPF Zero — 43,869 games**, trained for **43869 self-play games**. Architecture: `{"blocks": 2, "head": "flat", "kind": "resnet", "width": 32}`.
13
+
14
  Source training checkpoint SHA-256: `aa533bb8922231a3e0dc53dd2ff947cab6899ef09a9245463c89ac99ea776ef0`.
15
 
16
  Served `champion.pt` artifact SHA-256: `d5a7e3bda2966a79d22cf34f131976907260e679683d0e2680d935312f41c2f6`.
RESULTS.md CHANGED
@@ -1,8 +1,6 @@
1
  # Overnight evaluation — `20260909T235853Z`
2
 
3
- **Status:** `partial`
4
-
5
- > This experiment is incomplete or still running. Missing matches are shown as unplayed; no winner is inferred from partial coverage.
6
 
7
  ## Protocol
8
 
@@ -94,9 +92,12 @@ These intermediate continuation snapshots are included for audit. They do not by
94
 
95
  ## Promotion and publication
96
 
 
 
97
  - Promotion gate: `False`.
98
  - `duel_report`: [final-vs-served.json](final-vs-served.json)
99
  - Hugging Face: [wuschelschulz/gipf-zero](https://huggingface.co/wuschelschulz/gipf-zero)
 
100
 
101
  ## Raw duel reports
102
 
 
1
  # Overnight evaluation — `20260909T235853Z`
2
 
3
+ **Status:** `partial` (training/evaluation complete; publication recovered)
 
 
4
 
5
  ## Protocol
6
 
 
92
 
93
  ## Promotion and publication
94
 
95
+ > Training/evaluation complete; publication recovered. The raw runner status and errors are retained below for audit.
96
+ - Raw runner errors: huggingface-upload exited 1; github-push exited 128
97
  - Promotion gate: `False`.
98
  - `duel_report`: [final-vs-served.json](final-vs-served.json)
99
  - Hugging Face: [wuschelschulz/gipf-zero](https://huggingface.co/wuschelschulz/gipf-zero)
100
+ - Publication recovery receipt: [publication-recovery.json](publication-recovery.json) (`event=complete`).
101
 
102
  ## Raw duel reports
103
 
final-checkpoint-publication.json CHANGED
@@ -18,13 +18,21 @@
18
  },
19
  "games": 93916,
20
  "iteration": 2873,
21
- "evaluation_pending": true,
22
  "evaluation_report": "reports/overnight/20260909T235853Z/final-vs-served.json",
23
- "interpretation": "Frozen final training checkpoint only; evaluation and promotion were pending when this receipt was created. It is not claimed as strongest or served champion.",
24
  "upload_protocol": {
25
  "per_process_timeout_seconds": 180,
26
  "maximum_retries": 1,
27
  "network_upload_attempts": 1
28
  },
29
- "repository_commit": "8682c0c683117b697b3f1c36ff63008acf3295b8"
 
 
 
 
 
 
 
 
30
  }
 
18
  },
19
  "games": 93916,
20
  "iteration": 2873,
21
+ "evaluation_pending": false,
22
  "evaluation_report": "reports/overnight/20260909T235853Z/final-vs-served.json",
23
+ "interpretation": "Preserved final training checkpoint; completed evaluation did not pass promotion. The separately published champion.pt remains served.",
24
  "upload_protocol": {
25
  "per_process_timeout_seconds": 180,
26
  "maximum_retries": 1,
27
  "network_upload_attempts": 1
28
  },
29
+ "repository_commit": "8682c0c683117b697b3f1c36ff63008acf3295b8",
30
+ "final_evaluation": {
31
+ "wins": 38,
32
+ "losses": 42,
33
+ "cutoffs": 0,
34
+ "unfinished": 0,
35
+ "promotion_passed": false,
36
+ "report": "final-vs-served.json"
37
+ }
38
  }
publication-recovery.json ADDED
@@ -0,0 +1,26 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "time": 1789018647.126273,
3
+ "event": "complete",
4
+ "summary": "reports/overnight/20260909T235853Z/summary.json",
5
+ "hf_repo": "wuschelschulz/gipf-zero",
6
+ "report": "reports/overnight/20260909T235853Z/RESULTS.md",
7
+ "continuation_reports": [
8
+ "reports/overnight/20260909T235853Z/league-024737-duel.json",
9
+ "reports/overnight/20260909T235853Z/league-032337-duel.json",
10
+ "reports/overnight/20260909T235853Z/league-035932-duel.json",
11
+ "reports/overnight/20260909T235853Z/league-043512-duel.json"
12
+ ],
13
+ "recovery_reason": "Supervisor environment lacked working publication credentials. Reran bounded finalizer in authenticated interactive environment.",
14
+ "github_commit": "2379f2e",
15
+ "preserves_original_runner_errors": true,
16
+ "hf_verification": {
17
+ "revision": "55e998f5dde0c66ff6ca77eed73ba32fd797b068",
18
+ "served_artifact_sha256": "d5a7e3bda2966a79d22cf34f131976907260e679683d0e2680d935312f41c2f6",
19
+ "bytes": 4663985,
20
+ "download_hash_matches": true,
21
+ "cpu_load_and_finite_forward": true,
22
+ "comparison_checkpoints": 10,
23
+ "finalist_checkpoints": 1
24
+ },
25
+ "github_release": "https://github.com/wusche1/gipf-zero/releases/tag/overnight-20260909T235853Z"
26
+ }