Provides tools for inspecting, localizing, editing, and evaluating PyTorch model weights without training, including activation patching, circuit discovery, weight-editing operations, and gated evaluation of proposed edits.
This server does not currently have a deployable release on Glama. Follow the repository's setup instructions to install and configure it for your MCP client, including any required credentials and local dependencies.
Maintainers: quantomania-rudolph · Ethan Marreel (HIRO Industries) — ethan@hiroindustries.com · see CONTRIBUTORS.md
Weights Warden gives an LLM overseer (or any script) a full read → localize → edit → judge loop over a PyTorch model without training. Everything below is what the harness actually exposes through MCP and ./h call — see agent/TOOLS.md, agent/tool_spec.json, and engine/agent_tools/dispatch.py. Tool lists depend on HARNESS_MODE in harness.env (llm for transformers/MLPs via a ModelMap; default for the Nav2D policy demo).
The harness is built for agents that run their own science: orient with inventory / list_ops, form hypotheses (note), try edits for free (screen, sweep), spend a fixed budget of charged propose calls, and resume after context loss (state). Every tool call and verdict appends to a hash-chained ledger (runs/<run>/). The agent must not call finish — that single held-out evaluation is operator-only (./h finish).
Enforced accept rule on each propose (see AGENTS.md): success must rise by ≥3 points with paired-bootstrap CI lower bound > 0, in-distribution drop ≤3 points, and canary must pass. screen / sweep use separate research data and do not count as proof.
Optional mid-training hook (engine/agent_tools/train_hook.py): snapshot weights, hand a request dir to an agent, gate edits with your evaluator, auto-rollback if a watched metric regresses.
engine/any_model/patching.py)On clean/corrupt prompt pairs (LLM mode) or policy state batches (Nav mode), the stack separates:
Mechanism | What it is | Role |
Node attribution patching | First-order: ((a_\text{corrupt} - a_\text{clean}) \cdot \partial \text{metric}/\partial a) | Ranks heads / MLP units / hidden units |
Exact node patching | Re-run forward with corrupt activations inserted | Oracle for faithfulness and circuit scores |
Edge attribution (EAP) | Gradient at each reader input × writer contribution | Shortlists residual edges (transformers) |
Exact edge patching | Add ((w_\text{corrupt} - w_\text{clean})) at one reader only | Validates edges; used in |
Nav2D mode adds exact edge-removal over the full residual graph (circuit tool → engine/nav_circuits/eap_nav.py): memorisation vs expert contrast on writer→reader edges and neuron attributions (neuron_scan).
engine/any_model/llm_cli.py, circuit.py, circuit_grow.py)Tool | Grounded behaviour |
| Validates clean/corrupt JSON pairs (length, logit preference, corruption hurts). |
| Exact patching over every head and MLP unit; keep nodes with strong positive importance, t-stat ≥ 2, sign agreement across pairs; reports faithfulness ( |
| Starts from a selected circuit; greedy expansion (upstream writers via exact edge patching, high-importance nodes, optional baseline co-activation neighbours) until sufficiency ( |
| In-distribution activation map: dead / always-on units, co-activation clusters ( |
| Units correlated with a saved circuit; optional verification vs random units on sufficiency. |
| Multiple circuits: macro core across behaviours ( |
| Turns circuit nodes into |
A circuit that is necessary is not always sufficient — the README process in AGENTS.md expects grow before trusting edits.
LLM mode: inventory, weights (sliced raw matrices ≤4000 values; HARNESS_RAW_VIEW=0 = summary-only ablation arm), diff between checkpoints 0…9 / final.
Nav2D mode (engine/agent_tools/read_tools.py, analyst_tools.py):
Tool | What you get |
| Any matrix or parameter slice (biases, LayerNorm), with norms and values. |
| Per-head Wq / Wk / Wv / Wo norms, QK circuit ((W_q^\top W_k)) and OV circuit spectra, which input dims each projection reads, |
| Mean Q×K patterns by token type, entropy, example maps ( |
| MLP hidden units: train vs held-out firing stats; most-different units. |
| Declared |
| Label-free contrast attribution per MLP neuron (training labels vs expert on held-out). |
| Exact edge-level contrast on the Nav residual graph (~90s). |
| Per-module relative weight change and mean action change between checkpoints. |
| Free teacher-forced L1 shift toward expert — not closed-loop success. |
engine/core/ops.py + engine/agent_tools/more_ops.py)list_ops prints every op, argument ranges, and matrix shapes for this model. Registered set depends on mode (engine/agent_tools/run.py):
Heads & MLP units
scale_head — scale one attention head’s output projection columns (gamma=0 ablates).
head_temperature — scale query rows (attention logits; where the head looks).
scale_neurons / prune_neurons — scale or zero MLP hidden units (Nav: fc2 columns; LLM: mapped down_proj / MLP lin).
head_transplant — blend one head’s Q/K/V/O from another (Nav / full opt-in only).
scale_token_embed — scale token embedding linears (Nav policy).
Whole-matrix & SVD
project_out_top_svd, rank_truncate, shrink_subspace — remove or damp dominant singular directions of a named matrix.
revert_delta — move all weights toward an earlier checkpoint ckpt_000…009.
blend_matrix / blend_slice — partial checkpoint blend on one matrix or row/column subset.
Fine-grained value edits (caps relative to matrix RMS; see more_ops.py)
nudge_weights — add deltas to individual entries along one row or column.
scale_slice — multiply whole rows or columns by gamma.
bias_shift — nudge bias vector entries.
rank1_update — small structured ( \alpha , u v^\top ) patch on a sub-block.
Composition
composite — up to four sub-edits in one charged propose (validated independently).
LLM mode uses llm_ops.enable() plus matrix/value ops; Nav-only structured ops (e.g. head_transplant) are not registered when HARNESS_MODE=llm.
Tool | Cost | Data | Returns |
| Free (capped/run) | Research episodes | Δsuccess + CI — hint only |
| Free (2–9 strengths) | Research episodes | Curve + best value; each point uses a screen |
| Charged (budget default 6) | Selection set + ID/canary gates | Verdict, Δsuccess, CI, |
| Free | — | Budget left, ledger chain integrity |
| Free | — | Digest for resume: history, notes, best edit |
| Free | — |
|
Outputs can include a one-line caveat (engine/any_model/caveats.py) stating how far to trust the measurement.
Cursor / Claude Code: open the repo, pip install -r requirements.txt, tell the agent “Read AGENTS.md and follow it.” Enable the weight-editing-harness MCP server (.cursor/mcp.json / .mcp.json).
Any shell agent: ./h call TOOL '{"…"}' or agent/PROMPT.md + agent/tool_spec.json.
Operator: ./h finish for held-out read; ./h status anytime.
Not zero-config: implement a backend (evaluate, forward, probe_obs, load_ckpt) and set harness.env:
HARNESS_PYTHONPATH=/path/to/your/backend
HARNESS_BACKEND=my_backend:MyBackend
HARNESS_MODE=llm # transformers / MLPs via ModelMap + structure.pyRun engine/agent_tools/adoption_check.py on your model + backend before agent edits. For training integration, wire HarnessHook with your evaluator.
Path | Role |
| Agent process, tools, schemas |
| Ledgers, notes, saved circuits (local) |
| Loop, ops, patching, MCP server, backends |
| Backend, mode, budget, flags |
Toy models/, docs/, scripts/ ship in full releases when present in your tree.
Public evidence is toy-scale until you run on your checkpoint. Exact patching beats first-order edge scores for decisions. Neuron-level reads are weaker than head/block edits. Only accepted propose counts; only finish is final held-out.
MIT — see LICENSE.
This server cannot be deployed
LLM cost optimization gateway, repo token-waste auditor & model arbitrage MCP server.
Pre-execution governance for AI agents. Deterministic PASS/FAIL/REVIEW verdicts, replayable proof.
Ledger of record for AI spend: verifiable receipts, waste auditor & model arbitrage MCP.
See, price, and control every tool call your AI agents make: policy checks, cost, and audit tools.
