Oratis commited on
Commit
26e4326
·
verified ·
1 Parent(s): 038efa5

归档自有合成smoke输入与历史代码版本,保留逐成员映射和实验边界

Browse files
README.md CHANGED
@@ -36,3 +36,7 @@ The archived values are historical observations, including failures, inconsisten
36
  Gaming500-derived manifests and labels attribute [Markov AI's Gaming 500 Hours](https://huggingface.co/datasets/markov-ai/gaming-500-hours), whose current card declares [CC BY-NC 4.0](https://creativecommons.org/licenses/by-nc/4.0/). The current source revision observed during closeout was `0e84517002c33b030e0aff30c8afc7714a8fc8cb`; this is not a claim that all historical runs used that revision. Source benchmark images are not redistributed here. Keep attribution and noncommercial restrictions when using the corresponding derived material.
37
 
38
  Model families referenced by experiments include [Qwen3-VL](https://huggingface.co/Qwen/Qwen3-VL-4B-Instruct) and [DINOv2](https://github.com/facebookresearch/dinov2); their official source/model licenses are separate from rights in benchmark data or generated labels. Archived teacher identifiers and evaluation-only flags remain unchanged. The public [Diogenes source snapshot](https://github.com/CuddlerAI/Diogenes-Archive) documents available code and reconstruction limits. Read [NOTICE.md](NOTICE.md) before reuse.
 
 
 
 
 
36
  Gaming500-derived manifests and labels attribute [Markov AI's Gaming 500 Hours](https://huggingface.co/datasets/markov-ai/gaming-500-hours), whose current card declares [CC BY-NC 4.0](https://creativecommons.org/licenses/by-nc/4.0/). The current source revision observed during closeout was `0e84517002c33b030e0aff30c8afc7714a8fc8cb`; this is not a claim that all historical runs used that revision. Source benchmark images are not redistributed here. Keep attribution and noncommercial restrictions when using the corresponding derived material.
37
 
38
  Model families referenced by experiments include [Qwen3-VL](https://huggingface.co/Qwen/Qwen3-VL-4B-Instruct) and [DINOv2](https://github.com/facebookresearch/dinov2); their official source/model licenses are separate from rights in benchmark data or generated labels. Archived teacher identifiers and evaluation-only flags remain unchanged. The public [Diogenes source snapshot](https://github.com/CuddlerAI/Diogenes-Archive) documents available code and reconstruction limits. Read [NOTICE.md](NOTICE.md) before reuse.
39
+
40
+ ## Synthetic fixtures and additional historical code
41
+
42
+ [Procedural smoke fixtures and eight historical code snapshots](retired-training-source/synthetic-fixtures/README.md) preserve 144 logical synthetic NPZ/PNG files with exact aliases, fixed annotation links, and the original experimental limitations.
retired-training-source/synthetic-fixtures/MANIFEST.jsonl ADDED
The diff for this file is too large to render. See raw diff
 
retired-training-source/synthetic-fixtures/README.md ADDED
@@ -0,0 +1,46 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # Historical synthetic smoke fixtures and code snapshots
2
+
3
+ This supplement preserves project-generated synthetic development fixtures. The images are procedural color tiles with an action-controlled square and a distractor, with synthetic flash and vignette effects. These are not captures of a commercial game or evidence of real-game model performance.
4
+
5
+ `synthetic-fixtures.tar.gz` stores 120 unique, original-byte files: 72 NPZ frame arrays and 48 PNG images. `MANIFEST.jsonl` maps all 144 historical logical files (96 NPZ plus 48 PNG), including 24 exact NPZ aliases, to the unique payloads. Both retired `ml_code.tgz` source packages contain the same 144 logical payloads by name, size and SHA-256. The NPZ arrays have key `frames`, dtype `uint8`, and shape `[120, 96, 96, 3]`. They contain no Python objects or string arrays. PNG files have only IHDR, IDAT and IEND chunks. The normalized outer archive omits operating-system sidecars and original owner/path metadata; the NPZ and PNG bytes are unchanged.
6
+
7
+ The historical synthetic annotations and manifests are already preserved at the fixed revision below. Each NPZ mapping points to its original per-chunk labels, metadata and session metadata. The original `rights=synthetic`, `first_party=true`, and `is_trainable=true` fields are preserved; they describe this controlled synthetic task and do not certify production game data. The associated 400-step and 600-step synthetic IDM experiments used different gate policies and must not be interpreted as a direct real-game quality comparison.
8
+
9
+ `historical-code-extra.tar.gz` adds eight distinct historical versions of project source, tests and documentation, with all original bytes retained. Their historical claims, gates and operational examples describe the retired experiment; they are not current evaluation results or instructions to reactivate cloud resources. No dependencies, base model weights, tokenizer caches or credentials are included.
10
+
11
+ Synthetic data follows this research dataset's CC BY-NC 4.0 terms. Project source follows the separately archived project's Apache-2.0 license; references to upstream assets and historical dataset restrictions in original documentation remain applicable.
12
+
13
+ Validation: all original source containers and selected members were SHA-256 checked; the source gzip CRC and normalized archive member hashes were checked. Every NPZ was parsed with `allow_pickle=False` and numeric-only dtype checks, every PNG chunk CRC was checked, and all 48 PNGs plus first/middle/last frames for the 72 unique arrays were visually reviewed. This is representative frame inspection, not a claim that every frame was manually viewed.
14
+
15
+ ## Fixed annotation and generator references
16
+
17
+ {
18
+ "repo": "DiogenesLab/Diogenes_Vision_Language_Research_Archive_v01",
19
+ "revision": "038efa534795034d13a0a4f08cb5be2e053ceabe",
20
+ "generator": {
21
+ "repo": "DiogenesLab/Diogenes_Vision_Language_Research_Archive_v01",
22
+ "revision": "038efa534795034d13a0a4f08cb5be2e053ceabe",
23
+ "archive": "retired-training-source/files.tar.gz",
24
+ "member": "artifacts/2f24414d8c6b4562aaaf95215fe1d56544c200ccf61961dbc198cc457d07df87/synthetic.py",
25
+ "size": 9530,
26
+ "sha256": "2f24414d8c6b4562aaaf95215fe1d56544c200ccf61961dbc198cc457d07df87"
27
+ },
28
+ "bench_metadata": [
29
+ {
30
+ "repo": "DiogenesLab/Diogenes_Vision_Language_Research_Archive_v01",
31
+ "revision": "038efa534795034d13a0a4f08cb5be2e053ceabe",
32
+ "archive": "retired-training-source/files.tar.gz",
33
+ "member": "artifacts/a09750cb6e2e90ae38d29ae433ddca905be7abf7b7fc73b416b6f67d66496c13/bench_info.json",
34
+ "size": 155,
35
+ "sha256": "a09750cb6e2e90ae38d29ae433ddca905be7abf7b7fc73b416b6f67d66496c13"
36
+ },
37
+ {
38
+ "repo": "DiogenesLab/Diogenes_Vision_Language_Research_Archive_v01",
39
+ "revision": "038efa534795034d13a0a4f08cb5be2e053ceabe",
40
+ "archive": "retired-training-source/files.tar.gz",
41
+ "member": "artifacts/ec4066bb9ffb621b751d6d8c5a31f3a50bc99cbeed93b2ec26d86e9cccfc55c0/labels.jsonl",
42
+ "size": 24895,
43
+ "sha256": "ec4066bb9ffb621b751d6d8c5a31f3a50bc99cbeed93b2ec26d86e9cccfc55c0"
44
+ }
45
+ ]
46
+ }
retired-training-source/synthetic-fixtures/historical-code-extra.tar.gz ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:13e5954cef62322c8ca57a33d1eecb84dd8082d484efe65385eab2f7bb34f05c
3
+ size 34253
retired-training-source/synthetic-fixtures/synthetic-fixtures.tar.gz ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:9bdc065694b76f067b13adcba6cd1da54a5eebf7363ce01709d42f610463e4c4
3
+ size 3079818