SLWM-v2 β€” Study artifacts (Wave 7 spatial reference)

Study artifacts for reference β€” unpublished manuscript.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.

The 453M-parameter spatial-grid refinement used for unprivileged visual planning, beyond-window perturbed goal switching, and zero-shot camera-fault detection in Wave 7.

Research by Diogo Antunes, at valuealigned.ai and keplen.ai. Shared as study artifacts in the valueai-org organization.

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.

Study artifact reference

Property Value
Original run x1b-w7_slwm_450m_full-s2001
Experimental wave 7
Training seed 2001
Strict total parameters 453,199,938
Training steps 2,200
Batch size 12 sequences
Training sequence length 32 ticks
Latent width / layers / slots per tick 1152 / 18 / 24
Runtime/source reference cbb2b38

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 The unpublished manuscript reports three-seed cohorts; this repository contains one checkpoint and its own recorded metrics.

Wave 7 retains end-effector proprioception and uses a spatial visual readout and learned hazard grid instead of privileged wall coordinates. This package is the full, shielded seed-2001 reference. The archived configuration retains its original readout identifier; use the pinned Wave 7 implementation, which includes the 8 Γ— 8 grid functions.

Files and integrity

File Bytes SHA-256
weights_bf16.npz 719,945,456 fc694e6dbfb0239faff96113a3450f70e39721e5b791e09820ed1b45a0d25cb5
portable_int8.npz 415,131,879 2ffbb5755015c5ce45a2f4a4bd1fab01ff69d664b47da9c27d5e35372fb8a09e
  • weights_bf16.npz: the project's mixed BF16/FP32 export. Matrix BF16 values are stored as uint16 bit patterns, with a dtype manifest.
  • portable_int8.npz: the project's symmetric per-channel INT8 export, with scales and a quantization manifest.
  • config.json, receipt.json, eval_metrics.json, and train.jsonl: original run artifacts, preserved byte for byte.
  • artifact_manifest.json and SHA256SUMS: source mapping, artifact hashes, and export-validation results.
  • load_checkpoint.py: a small, pickle-free NumPy loader for both formats.

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.

The provided loader expands weights to float32. INT8 file size is not the runtime memory requirement, and this loader does not perform INT8 matrix computation. Model weights, activations, and planning buffers are separate from the paper's bounded streaming-state figures.

Download and load

Install numpy and huggingface_hub in your Python environment, then:

import json
import sys
from pathlib import Path
from huggingface_hub import snapshot_download

folder = Path(snapshot_download(
    repo_id="valueai-org/SLWM-v2-453M",
    allow_patterns=["portable_int8.npz", "config.json", "receipt.json", "load_checkpoint.py"],
))
sys.path.insert(0, str(folder))
from load_checkpoint import load_checkpoint

receipt = json.loads((folder / "receipt.json").read_text())
params = load_checkpoint(
    folder / "portable_int8.npz",
    expected_sha256=receipt["portable_int8_sha256"],
)
assert sum(value.size for value in params.values()) == receipt["parameter_counts"]["strict_total"]

For the BF16 export, download weights_bf16.npz instead and use receipt["weights_bf16_sha256"] for verification. params is a dictionary of NumPy float32 arrays. JAX execution requires JAX and the appropriate modules from the pinned research source; follow that source for input preparation, modality masks, state updates, and planning.

Use models/x1b_wave6_core.py for the recurrent forward pass and models/x1b_wave7_core.py for the spatial visual, hazard, goal, and world-head readouts. The matching evaluation entry points are in evals/x1b_eval_wave7.py. The older Wave 4 replay CLI alone does not exercise the Wave 7 spatial planner.

Run an example and validation checks

Use the runnable examples and validation-set guide from the SLWM source repository. It explains the development and held-out data splits, how to verify local data-bank checksums, and which recorded inputs are not included with these checkpoints.

From a current checkout of the source repository, after installing the guide's Python dependencies:

python scripts/eval/x1b_study_examples.py \
  --mode example --repo-id valueai-org/SLWM-v2-453M --format int8 \
  --output /tmp/slwm-v2-453m-example.json

This is a generated, deterministic functional example using the selected trained checkpoint; it is not a benchmark score. The guide provides BF16 commands for the original evaluation functions and the exact bank paths and split names for validation.

Recorded results for this checkpoint

These values come directly from this run's receipt.json. They are single-seed results, not the manuscript's cohort mean Β± standard deviation. The upload does not claim that the quantized export has received an additional independent evaluation.

Recorded metric Seed 2001
T = 160 perturbed goal-switch success 87.5000 %
Unseen-layout subgoal MPPI success 51.5625 %
Matched explicit-latch reactive success 31.2500 %
Dynamic-visual associative recall at T = 1,024 6.2500 %
Unseen camera-corruption AUROC 99.9057 %
Held-out-scene xArm6 four-step error 31.4589 mm

The original eval_metrics.json provides the wider evaluation record, including failures. Compare architectures within the same wave and observation/training setting.

Scope and limitations

Dense spatial learning coexists with weak high-cardinality synthetic recall in this wave. Strong goal recovery does not imply perfect associative memory. Extreme dynamics shifts remain a failure case, held-out spoken-instruction discrimination is close to chance, and robot evaluation is offline. AUROC describes ranking of the tested faults, not a deployment threshold or a general safety guarantee.

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.

Sources and licensing

These study artifacts, including the BF16 and INT8 weights, use the MIT license, as specified by the author. The complete LICENSE and SOURCE_LICENSE files match the corrected GitHub project license. The copyright owners are Diogo Antunes (contact@diogoantunes.ai), ValueAI (contact@valuealigned.ai), and Keplen (contact@keplen.ai).

Citation

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.

Downloads last month
46
Inference Providers NEW
This model isn't deployed by any Inference Provider. πŸ™‹ Ask for provider support