diogoantunes commited on
Commit
e8f249e
·
verified ·
1 Parent(s): 5f24c75

Label unpublished-paper study artifacts and apply the GitHub MIT license

Browse files
LICENSE ADDED
@@ -0,0 +1,25 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ MIT License
2
+
3
+ Copyright (c) 2026 Diogo Antunes <contact@diogoantunes.ai>
4
+ Copyright (c) 2026 ValueAI <contact@valuealigned.ai>
5
+ Copyright (c) 2026 Vaisys <contact@vaisys.com>
6
+ Copyright (c) 2026 Splendor <contact@splendor-os.org>
7
+ Copyright (c) 2026 UniversoAI <contact@universo-ai.com>
8
+
9
+ Permission is hereby granted, free of charge, to any person obtaining a copy
10
+ of this software and associated documentation files (the "Software"), to deal
11
+ in the Software without restriction, including without limitation the rights
12
+ to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
13
+ copies of the Software, and to permit persons to whom the Software is
14
+ furnished to do so, subject to the following conditions:
15
+
16
+ The above copyright notice and this permission notice shall be included in all
17
+ copies or substantial portions of the Software.
18
+
19
+ THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
20
+ IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
21
+ FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
22
+ AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
23
+ LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
24
+ OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
25
+ SOFTWARE.
README.md CHANGED
@@ -1,4 +1,5 @@
1
  ---
 
2
  language:
3
  - en
4
  library_name: jax
@@ -8,18 +9,22 @@ tags:
8
  - working-memory
9
  - embodied-control
10
  - research
 
 
11
  - npz
12
  ---
13
 
14
- # SLWM-X450M — Wave 6 memory reference
 
 
15
 
16
  The full 453M-parameter, Epps-Pulley SIGReg reference from the single-factor ablation study. It preserves key shielding, the sensory bypass, and decoupled readout heads.
17
 
18
- Research by **Diogo Antunes**, at [valuealigned.ai](https://valuealigned.ai) and [keplen.ai](https://keplen.ai). Published in the **valueai-org** organization.
19
 
20
  This is an original, unmodified checkpoint from the SLWM research archive. It uses a **custom NumPy NPZ format** and the SLWM JAX runtime; it is not a Transformers `AutoModel` checkpoint or a chat model.
21
 
22
- ## Exact release
23
 
24
  | Property | Value |
25
  |---|---|
@@ -33,7 +38,7 @@ This is an original, unmodified checkpoint from the SLWM research archive. It us
33
  | Latent width / layers / slots per tick | 1152 / 18 / 24 |
34
  | Runtime/source reference | [`cbb2b38`](https://github.com/ValueAlignedAI/slwm/tree/cbb2b38794359adfefe07b19e9b7caa113b25b47) |
35
 
36
- Seed 2001 is released as a fixed reference, not selected by ranking test scores. The research paper reports three-seed cohorts; this repository contains one checkpoint and its own recorded metrics.
37
 
38
  This is the Wave 6 full model with end-effector-only proprioception. It is useful as a reference for high-cardinality dynamic-visual overwrite memory and the global-pooling spatial bottleneck. It is not the Wave 7 spatial refinement.
39
 
@@ -47,7 +52,7 @@ This is the Wave 6 full model with end-effector-only proprioception. It is usefu
47
  - `weights_bf16.npz`: the project's mixed BF16/FP32 export. Matrix BF16 values are stored as uint16 bit patterns, with a dtype manifest.
48
  - `portable_int8.npz`: the project's symmetric per-channel INT8 export, with scales and a quantization manifest.
49
  - `config.json`, `receipt.json`, `eval_metrics.json`, and `train.jsonl`: original run artifacts, preserved byte for byte.
50
- - `release_manifest.json` and `SHA256SUMS`: source mapping, artifact hashes, and export-validation results.
51
  - `load_checkpoint.py`: a small, pickle-free NumPy loader for both formats.
52
 
53
  All original files were checked against their recorded SHA-256 values before upload. Both weight exports were decoded on CPU, checked for finite values, and counted against the strict total and component counts. These are packaging/integrity checks, not new benchmark evaluations.
@@ -100,17 +105,17 @@ The original `eval_metrics.json` provides the wider evaluation record, including
100
 
101
  The strong associative-recall result coexists with weak unprivileged visual planning and near-chance unseen camera-fault detection in this wave. Do not substitute this checkpoint for SLWM-v2 when evaluating the later spatial-grid planner. Physical xArm6 results remain offline recorded-trajectory evaluations.
102
 
103
- The checkpoint is intended for research and reproducibility. The upload contains no raw training recordings, datasets, optimizer state, or private server access configuration. The pinned source commit is a public runtime/release reference, not a separately verified training-code commit: the archived receipts do not record the training Git SHA.
104
 
105
  ## Sources and licensing
106
 
107
- - [Original manuscript v0.3 (PDF)](https://diogoantunes.ai/media/publications/slwm_x1b_paper_v03.pdf)
108
  - [Expanded article, figures, and experiment context](https://diogoantunes.ai/publications/signal-latent-world-models)
109
  - [SLWM source repository](https://github.com/ValueAlignedAI/slwm/tree/cbb2b38794359adfefe07b19e9b7caa113b25b47)
110
  - [Original experiment records](https://github.com/ValueAlignedAI/slwm/tree/cbb2b38794359adfefe07b19e9b7caa113b25b47/experiments/multimodal/x1b/spatiotemporal-jepa-1b-20261111/runs_wave6/x1b-w6_slwm_450m_full_epps_sigreg-s2001)
111
 
112
- The source code is distributed under the MIT license reproduced in `SOURCE_LICENSE`. No separate model-weight license was specified in the archived release metadata; this upload does not invent an additional license grant.
113
 
114
  ## Citation
115
 
116
- Diogo Antunes. *Signal-Latent World Models at 1.008B Scale: Separating Fast Sensory Grounding from Editable Working Memory Under O(1) Streaming Control.* Preprint, manuscript v0.3, November 14, 2026.
 
1
  ---
2
+ license: mit
3
  language:
4
  - en
5
  library_name: jax
 
9
  - working-memory
10
  - embodied-control
11
  - research
12
+ - study-artifacts
13
+ - unpublished-manuscript
14
  - npz
15
  ---
16
 
17
+ # SLWM-X450M — Study artifacts (Wave 6 memory reference)
18
+
19
+ > **Study artifacts for reference — unpublished manuscript.** These checkpoints accompany the manuscript draft v0.3, which has not yet been published. They are shared for inspection and reproducibility of the study, not as a formal model release.
20
 
21
  The full 453M-parameter, Epps-Pulley SIGReg reference from the single-factor ablation study. It preserves key shielding, the sensory bypass, and decoupled readout heads.
22
 
23
+ Research by **Diogo Antunes**, at [valuealigned.ai](https://valuealigned.ai) and [keplen.ai](https://keplen.ai). Shared as study artifacts in the **valueai-org** organization.
24
 
25
  This is an original, unmodified checkpoint from the SLWM research archive. It uses a **custom NumPy NPZ format** and the SLWM JAX runtime; it is not a Transformers `AutoModel` checkpoint or a chat model.
26
 
27
+ ## Study artifact reference
28
 
29
  | Property | Value |
30
  |---|---|
 
38
  | Latent width / layers / slots per tick | 1152 / 18 / 24 |
39
  | Runtime/source reference | [`cbb2b38`](https://github.com/ValueAlignedAI/slwm/tree/cbb2b38794359adfefe07b19e9b7caa113b25b47) |
40
 
41
+ Seed 2001 is provided as a fixed reference, not selected by ranking test scores. The unpublished manuscript reports three-seed cohorts; this repository contains one checkpoint and its own recorded metrics.
42
 
43
  This is the Wave 6 full model with end-effector-only proprioception. It is useful as a reference for high-cardinality dynamic-visual overwrite memory and the global-pooling spatial bottleneck. It is not the Wave 7 spatial refinement.
44
 
 
52
  - `weights_bf16.npz`: the project's mixed BF16/FP32 export. Matrix BF16 values are stored as uint16 bit patterns, with a dtype manifest.
53
  - `portable_int8.npz`: the project's symmetric per-channel INT8 export, with scales and a quantization manifest.
54
  - `config.json`, `receipt.json`, `eval_metrics.json`, and `train.jsonl`: original run artifacts, preserved byte for byte.
55
+ - `artifact_manifest.json` and `SHA256SUMS`: source mapping, artifact hashes, and export-validation results.
56
  - `load_checkpoint.py`: a small, pickle-free NumPy loader for both formats.
57
 
58
  All original files were checked against their recorded SHA-256 values before upload. Both weight exports were decoded on CPU, checked for finite values, and counted against the strict total and component counts. These are packaging/integrity checks, not new benchmark evaluations.
 
105
 
106
  The strong associative-recall result coexists with weak unprivileged visual planning and near-chance unseen camera-fault detection in this wave. Do not substitute this checkpoint for SLWM-v2 when evaluating the later spatial-grid planner. Physical xArm6 results remain offline recorded-trajectory evaluations.
107
 
108
+ The checkpoint is intended for research and reproducibility. The upload contains no raw training recordings, datasets, optimizer state, or private server access configuration. The pinned source commit is a public runtime/study reference, not a separately verified training-code commit: the archived receipts do not record the training Git SHA.
109
 
110
  ## Sources and licensing
111
 
112
+ - [Unpublished manuscript draft v0.3 (PDF)](https://diogoantunes.ai/media/publications/slwm_x1b_paper_v03.pdf)
113
  - [Expanded article, figures, and experiment context](https://diogoantunes.ai/publications/signal-latent-world-models)
114
  - [SLWM source repository](https://github.com/ValueAlignedAI/slwm/tree/cbb2b38794359adfefe07b19e9b7caa113b25b47)
115
  - [Original experiment records](https://github.com/ValueAlignedAI/slwm/tree/cbb2b38794359adfefe07b19e9b7caa113b25b47/experiments/multimodal/x1b/spatiotemporal-jepa-1b-20261111/runs_wave6/x1b-w6_slwm_450m_full_epps_sigreg-s2001)
116
 
117
+ These study artifacts, including the BF16 and INT8 weights, use the **MIT license**, as specified by the author. The complete `LICENSE` file is copied unchanged from the [GitHub project](https://github.com/ValueAlignedAI/slwm/blob/cbb2b38794359adfefe07b19e9b7caa113b25b47/LICENSE). `SOURCE_LICENSE` retains the same original source-code notice.
118
 
119
  ## Citation
120
 
121
+ Diogo Antunes. *Signal-Latent World Models at 1.008B Scale: Separating Fast Sensory Grounding from Editable Working Memory Under O(1) Streaming Control.* Unpublished manuscript, draft v0.3, dated November 14, 2026.
SHA256SUMS CHANGED
@@ -1,11 +1,12 @@
1
- 6969e36d1c9470c75acf094589bd22fa470e033447f22acc65615cc9541f9fb0 README.md
 
2
  c56a80b248a6b638e49a4365fff816160532b95bb5264a47f9dc67ae8fe64a40 SOURCE_LICENSE
 
3
  5500e8e88155306471bda21e7453f8289cd5c60664b8adabaf8402df578f00ec config.json
4
  5de7a606143688034badb7ee4404c526e49cd2195094c07dff061857d6a93c43 eval_metrics.json
5
  27686dfc84db3c70b74999e2ee665df6833b99e380df93d3573c857c9c240ca8 load_checkpoint.py
6
  36d94dfe6db11a0d06b73f816bfd5a32c17b29865ab58ca4b778739dfb6a0e89 portable_int8.npz
7
  8226214c0be46e556fb86ad0142039a76f3df3011ae6952b37d0246b0ad7cd8b receipt.json
8
- 569456e59e5bb018a53e4fc7855aa80bd753588f52e3541defb83ecc82bb8000 release_manifest.json
9
  c06d962ea090965c6958b75beffae8b1a8c52ae1a45a35d5b5ab04da0de90c9b requirements.txt
10
  9ae25544ca459917e53e5e91f6871d10be93edd4cf7b1a444dd4b8babbabfffe train.jsonl
11
  fe63862af1a8d3440b0a8b1e580b4961a179c9a65b7da0957bf25c434dab45a8 weights_bf16.npz
 
1
+ c56a80b248a6b638e49a4365fff816160532b95bb5264a47f9dc67ae8fe64a40 LICENSE
2
+ f1b0ac83ca63fb37d6be20d302ffae386ad72eb1738ff9d3c8c5090e78ac2fc2 README.md
3
  c56a80b248a6b638e49a4365fff816160532b95bb5264a47f9dc67ae8fe64a40 SOURCE_LICENSE
4
+ cadd1eddcb5da125905b47da95da03c6319dcf3fd6fbf77d782adfa6fefa6804 artifact_manifest.json
5
  5500e8e88155306471bda21e7453f8289cd5c60664b8adabaf8402df578f00ec config.json
6
  5de7a606143688034badb7ee4404c526e49cd2195094c07dff061857d6a93c43 eval_metrics.json
7
  27686dfc84db3c70b74999e2ee665df6833b99e380df93d3573c857c9c240ca8 load_checkpoint.py
8
  36d94dfe6db11a0d06b73f816bfd5a32c17b29865ab58ca4b778739dfb6a0e89 portable_int8.npz
9
  8226214c0be46e556fb86ad0142039a76f3df3011ae6952b37d0246b0ad7cd8b receipt.json
 
10
  c06d962ea090965c6958b75beffae8b1a8c52ae1a45a35d5b5ab04da0de90c9b requirements.txt
11
  9ae25544ca459917e53e5e91f6871d10be93edd4cf7b1a444dd4b8babbabfffe train.jsonl
12
  fe63862af1a8d3440b0a8b1e580b4961a179c9a65b7da0957bf25c434dab45a8 weights_bf16.npz
release_manifest.json → artifact_manifest.json RENAMED
@@ -1,6 +1,6 @@
1
  {
2
  "repo_id": "valueai-org/SLWM-X450M-Memory",
3
- "title": "SLWM-X450M \u2014 Wave 6 memory reference",
4
  "wave": 6,
5
  "run_tag": "x1b-w6_slwm_450m_full_epps_sigreg-s2001",
6
  "seed": 2001,
@@ -59,5 +59,12 @@
59
  "all_finite": true
60
  }
61
  },
62
- "metrics_scope": "Original single-seed receipt; no new model-quality evaluation performed."
 
 
 
 
 
 
 
63
  }
 
1
  {
2
  "repo_id": "valueai-org/SLWM-X450M-Memory",
3
+ "title": "SLWM-X450M \u2014 Study artifacts (Wave 6 memory reference)",
4
  "wave": 6,
5
  "run_tag": "x1b-w6_slwm_450m_full_epps_sigreg-s2001",
6
  "seed": 2001,
 
59
  "all_finite": true
60
  }
61
  },
62
+ "metrics_scope": "Original single-seed receipt; no new model-quality evaluation performed.",
63
+ "artifact_kind": "study_reference_checkpoint",
64
+ "manuscript_status": "unpublished",
65
+ "formal_model_release": false,
66
+ "purpose": "Reference artifacts accompanying the unpublished manuscript draft v0.3.",
67
+ "license": "mit",
68
+ "license_source": "https://github.com/ValueAlignedAI/slwm/blob/cbb2b38794359adfefe07b19e9b7caa113b25b47/LICENSE",
69
+ "license_sha256": "c56a80b248a6b638e49a4365fff816160532b95bb5264a47f9dc67ae8fe64a40"
70
  }