Claude Skills by VectorSpaceLab

Evaluate LoRAPrune recovery metrics while verifying structured-mask LoRA merge compatibility and mechanism checks.

Run LoRAPrune's iterative LoRA optimization, moving-average importance update, and progressive structured mask schedule.

Run a bounded mechanism-faithful recovery experiment for movement pruning without reading the original source repository.

Update movement-pruning importance scores from weight gradients and diagnose whether weights move away from zero.

Compute deterministic score-based pruning masks using top-v or threshold rules for movement-pruning style experiments.

Compute soft movement-pruning sigmoid sparsity penalties and gradients for threshold-mask recovery checks.

Combine supervised, output distillation, and RAIL intermediate losses with validated RAIL-KD lambda weights.

Run a bounded mechanism-faithful RAIL-KD proxy experiment with executable recovery and validation evidence.

Compute RAIL-KD layerwise and concatenated intermediate representation losses from pooled hidden states.

Sample and audit RAIL-KD random teacher-to-student intermediate layer mappings for bounded knowledge distillation experiments.

Rearrange binary pruning masks using block-diagonal Fisher interactions while preserving per-layer cardinality.

Select Transformer heads and filters to prune using diagonal Fisher scores under a FLOPs budget.

Tune surviving pruning mask coefficients with damped least squares for activation reconstruction.

Compose generated pruning modules into an auditable no-retraining recovery experiment with mechanism checks.

Estimate per-channel activation norms for Wanda outlier-aware pruning from calibration tensors.

Evaluate Wanda pruning masks for sparsity, no-update invariants, and linear-output reconstruction error.

Build deterministic N:M semi-structured pruning masks from Wanda or other importance scores.

Compute row-wise Wanda pruning scores, masks, and pruned weights without any weight update.

Evaluate equation-6 atomic APT categorical probabilities and density-ratio losses for finite parameter atom sets.

Compute APT posterior-to-proposal-posterior transformations for Gaussian and finite normalized density settings without importance weighting.

Build bounded APT recovery experiments with executable evidence, generated-skill invocation logs, source manifests, and mechanism checks.

Run APT-style sequential simulator proposal rounds with auditable prior, proposal, observation, and update records for likelihood-free inference recovery.

Run bounded SBI benchmark recovery from generated skills while enforcing source boundaries and validator-compatible evidence.

Fit compact posterior approximations from SBI simulator pairs and emit posterior samples with optimizer traces.

Compute deterministic posterior two-sample diagnostics including C2ST-style accuracy for SBI benchmark recovery.

Build and validate simulation-based inference benchmark task records with priors, simulators, observations, and reference posteriors.

Run annealed Langevin dynamics with composed scores for F-NPSE or PF-NPSE posterior sampling.

Train and validate conditional denoising score models for F-NPSE or PF-NPSE simulation-based inference recoveries.

Compose F-NPSE and PF-NPSE posterior scores from learned score terms and prior corrections for SBI.

Run a bounded Gaussian/Gaussian SBI proxy recovery that exercises F-NPSE score training, composition, and sampling.

Execute a bounded reduced recovery for binary-tree Gaussian BNs using NaMI inverse contracts and a real standard-library optimizer step.

Build and validate ordered inverse factorization contracts for amortized inference networks from generated inverse graph parents.

Audit inverse Bayesian-network structures for naturalness, local I-map consistency, and edge-minimality evidence using d-separation checks.

Generate natural minimally faithful inverse graph structures from Bayesian-network DAGs using the paper's NaMI variable-elimination procedure.

Build graph-derived sparse attention masks for Graphically Structured Diffusion Model style recovery experiments and checks.

Run a bounded BCMF proxy recovery that exercises generated Graphically Structured Diffusion Model skills.

Encode mixed continuous and categorical variables and compute GSDM-style conditional diffusion denoising losses.

Check mask-preserving permutations and derive embedding groups for graphically structured diffusion models.

Train or proxy-train SBI density estimator families on simulator-generated parameter observation pairs.

Provide observation-conditioned posterior sampling and log-probability APIs for SBI recovery estimators.

Validate reduced or full SBI recovery artifacts, source boundaries, mechanism checks, and metric reporting.

Validate simulator, prior, observation, and simulation-record contracts for simulation-based inference recovery workflows.

Define Score SDE VE, VP, and sub-VP kernels with reverse and probability-flow drift checks.

Compute Score SDE probability-flow drift identities and compact likelihood accounting diagnostics.

Run reduced reverse-time SDE, probability-flow, and predictor-corrector sampling updates for Score SDE.

Build continuous denoising score-matching targets, losses, and reduced optimizer evidence for Score SDE.

Run a bounded mechanism-faithful proxy recovery for Black Box Variational Inference using generated BBVI estimator and optimizer skills.

Compute black-box variational inference score-function ELBO gradient estimates from samples, log-density values, and variational score functions.

Apply stochastic ascent and AdaGrad updates to black-box variational inference gradient estimates while recording parameter-change evidence.

Compute BBVI Rao-Blackwellized and score-function control-variate gradient estimators with variance diagnostics.