You can deploy this model and start earning money today!

Divinci-AI/
gpt-oss-120b-vindex

Name: gpt-oss-120b-vindex
Availability: InStock

larql

vindex

mechanistic-interpretability

Views

No views yet

GPT-OSS-120b (MoE) — vIndex

Source model: openai/gpt-oss-120b Vindex short ID: 02d09974 Layers: 36 Hidden size: 2880 Features per layer: 64

What This Is

A vindex (vector index) — a compact binary representation of the feature geometry of openai/gpt-oss-120b. It contains the top-64 SVD directions of every MLP gate_proj and down_proj matrix in the network, plus token embeddings, layer norms, and vocabulary projection metadata.

What This Is NOT

This is not a model you can run for inference. It has no weights sufficient to generate text. It is a mechanistic interpretability artifact: a feature database for probing, editing, and comparing what openai/gpt-oss-120b has learned.

Universal Constants (Phase 2 Measurements)

Measured via forward-pass hooks on a 256-token factual probe text.

Constant	Symbol	Value	Interpretation
FFN Sparsity	C1	N/A	Fraction of near-zero SwiGLU activations
Top-8 Prob Mass	C2	N/A	Probability mass on top-8 output tokens
Gate Coherence	C3	N/A	Mean cosine sim of adjacent gate_proj directions
Layer Temperature	C4	N/A	Mean per-neuron SwiGLU activation variance
Circuit Stages	C5	N/A	CKA transition count + 1

Notes: MoE architecture — Phase 2 skipped (OOM on A100). S[0] singular values show exponential growth with depth (L0=111 → L13=1630), opposite to Bonsai's flat spectrum. Gate 3 blocked: uniform routing weights make sparse feature attribution meaningless without expert calibration.

Gate 3 Status (DELETE Patch Test)

BLOCKED — uniform routing weights; needs expert calibration before Gate 3.

Gate 3 tests whether a rank-1 ΔW patch to gate_proj.weight at the top Paris→capital feature layer suppresses P(Paris) by ≥70% with ≤30% Berlin collateral damage.

Files

File	Description
`gate_vectors.bin`	Top-64 SVD directions of gate_proj per layer [L×F×H, f16]
`down_features.bin`	Top-64 SVD directions of down_proj per layer [L×F×H, f16]
`embeddings.bin`	Token embedding matrix [V×H, f16]
`norms.bin`	Layer norm weight vectors
`down_meta.bin`	Per-feature top-k vocabulary projections
`index.json`	Vindex metadata (layers, hidden_size, num_feats, etc.)
`manifest.json`	Build provenance (source SHA, extraction timestamp)
`SHA256SUMS`	File integrity checksums

How to Use

python

import numpy as np, json

vindex_dir = "path/to/downloaded/vindex"

with open(f"{vindex_dir}/index.json") as f:
    idx = json.load(f)

L, F, H = idx["num_layers"], idx["num_feats"], idx["hidden_size"]
V = idx["vocab_size"]

# Load gate feature directions [L, F, H]
gate = np.frombuffer(
    open(f"{vindex_dir}/gate_vectors.bin", "rb").read(),
    dtype=np.float16
).reshape(L, F, H).astype(np.float32)

# Load embeddings [V, H]
emb = np.frombuffer(
    open(f"{vindex_dir}/embeddings.bin", "rb").read(),
    dtype=np.float16
).reshape(V, H).astype(np.float32)

# Score a token against all features (cosine similarity)
emb_n = emb / (np.linalg.norm(emb, axis=1, keepdims=True) + 1e-8)
gate_n = gate / (np.linalg.norm(gate, axis=2, keepdims=True) + 1e-8)

token_id = 12379  # e.g., " Paris"
scores = gate_n @ emb_n[token_id]  # [L, F]
l_max, f_max = np.unravel_index(scores.argmax(), scores.shape)
print(f"Top feature: layer={l_max}, feature={f_max}, score={scores[l_max, f_max]:.4f}")

License

CC-BY-NC 4.0 — same terms as the source model. Research use only.

Citation

If you use this vindex in published work, please cite:

@misc{divinci2026vindex,
  title  = {vIndex: GPT-OSS-120b (MoE)},
  author = {Divinci AI},
  year   = {2026},
  url    = {https://huggingface.co/Divinci-AI/gpt-oss-120b-vindex}
}