Expand description
LLM Testing Integration for CLI
Simple LLM model testing functionality for the CLI.
Structs§
- Explore
Candidate Result - One explore-row: layout (+ optional toggle label) → measured decode-proxy tok/s + coherence.
- Explore
Report - Test
Result - Test result
Functions§
- benchmark_
model - CLI command to benchmark a single model
- list_
models - CLI command to list available models
- parse_
explore_ 🔒layouts - run_
activate_ qwen4exp_ native - Cold-activate the native C:/E: package layout. This is deliberately separate from generation: it proves the model can be opened with a bounded metadata read before a decode runner is allowed to touch weights.
- run_
app_ profile - Show / set application profile (interactive | live-fast | batch).
- run_
benchmark_ models - CLI command to benchmark models
- run_
comprehensive_ llm_ test - Comprehensive LLM capability test (load + inference in one session)
- run_
convert_ gguf_ to_ p64 - Convert a GGUF import file to native
.p64+ canonical.q42model metadata. - run_
cuda_ tc_ microbench - Dense CUDA WMMA microbench (mode-independent; reports whether TC path is live).
- run_
decode_ proxy - Short resident decode for passport child processes. Machine-readable line on stdout.
- run_
explore_ pipeline - Phase-0 explorer: convert missing layouts if source is GGUF, measure decode-proxy, rank, write JSON.
- run_
gather_ qwen4exp_ ple - Compute the checkpoint’s PLE n-gram row IDs from decode-order token IDs, then gather exactly those rows from the C:-resident table. This is an execution seam, not a text-generation benchmark: it proves that runtime token history selects and consumes trained PLE rows without mapping the table or using a synthetic hash.
- run_
generate_ report - CLI command to generate test report
- run_
gpu_ capability_ campaign - Probe which native GPU tiers this host has, measure the backend × mode decode matrix
in child processes, and write
machine-gpu-profile.json+apply-machine-gpu.ps1. - run_
ground_ check - Quant-graph dry-run: ground a prompt+answer pair (forces mode logic via unconditional ground).
- run_
hardware_ passport - Probe / load HardwarePassport and print the ranked circuit matrix.
- run_
inference_ mode - Print or set multi-mode inference approach (portable / cuda / quant-graph).
- run_lab
- Inference superiority lab instruments.
- run_
optimize_ pipeline - Remarkable one-shot: optional passport + auto-layout convert + activate knobs.
- run_
path_ select - Print (and optionally apply) the device-optimal inference path plan.
- run_
prepare_ qwen4exp_ native - Extract the Qwen4Exp PLE table to NVMe and create its compact HMC contract. The source GGUF stays in place; only its row-addressable PLE span is copied.
- run_
probe_ qwen4exp_ expert_ tile - Exercise a promoted C: expert tile using a true E: token embedding. This is one cached gate-plane operator, not a complete MoE or text decode.
- run_
probe_ qwen4exp_ gdn - Execute the trained recurrent token mixer with its real learned state. The two state vectors are explicitly bounded and live only for this CLI invocation; no model payload is copied from E: or C: into them.
- run_
probe_ qwen4exp_ hyper - Execute one trained attention- or FFN-side Hyper-Connection mix using the source checkpoint’s own weight rows. This is a forward component probe; it deliberately does not substitute an attention/GDN or MoE result.
- run_
probe_ qwen4exp_ moe - Execute one true Qwen4Exp routed-MoE operator. This is an operator validation seam, not text generation: it uses a token embedding directly instead of a fully evolved preceding token-mixer state.
- run_
probe_ qwen4exp_ ple - Exercise the actual C:-resident PLE reader without mapping the PLE table, opening the source GGUF, or allocating a row dynamically.
- run_
probe_ qwen4exp_ ple_ block - Execute the complete trained PLE residual insertion for one token. This is a physical C:/E: forward component check, not token generation: all later GDN/QSA, four-stream and MoE layers remain to be executed.
- run_
probe_ qwen4exp_ trunk - Verify the external trunk reader against the exact source identified by an
HMC. Only the bounded GGUF metadata prefix is read; the embedding row is
consumed through
TrunkNvmeReaderfrom E:. - run_
promote_ qwen4exp_ expert - Explicit cold-path promotion of one selected expert. This is deliberately not part of ordinary activation: the caller must name a C: cache directory and budget before any new artifact is made.
- run_
seed_ grounding - Seed quant-graph facts from bundled TSV / QUALIA_GROUNDING_FACTS.
- run_
test_ models - CLI command to run comprehensive LLM model tests
- run_
validate_ models - CLI command to validate models
- run_
verify_ qwen4exp_ native - Audit an existing Qwen4Exp HMC’s external source before activation.
- test_
single_ 🔒model - Test a single model
- validate_
model - CLI command to validate model structure