TensionLM-117M-TS-Reasoner-v6

This is the paraphrase-aware CPU TS reasoner for the frozen TensionLM-117M-Reasoning-v2 substrate.

v6 adds paraphrase pressure over v5:

• graph relations expressed as pipelines, handoffs, primary edges, and audit trails,
• arithmetic phrased as ledger/recipe/token/quotient/remainder traces,
• code phrased as augmented assignment, even-loop accumulation, branch, dict mutation, and list pop traces.

The point is still no-GPU reasoning: frozen language substrate plus explicit TS graph/program operators.

Eval receipts

Fixed benchmark scores:

System	TAC v2	TAC v3	TAC v4
TS-Reasoner-v6	120/120	120/120	120/120

Generated benchmark scores:

Distribution	Engine	Score
TAC-GEN standard, seeds 9001-9004	v5	12000/12000
TAC-GEN paraphrase, seed 9101 n=100/category	v5	0/300
TAC-GEN paraphrase, seeds 9101-9104	v6	12000/12000

These are system scores, not raw LLM scores.

Usage

python inference.py --prompt "Handoff log: a hands off to b; b hands off to c. Ignore the separate handoff x hands off to b. The handoff chain beginning at a ends at" --category transitivity --show_trace

python inference.py --prompt "Ledger entry: begin 20, pay out 5, add back 3, scale by 2. The ledger closes at" --category arithmetic --show_trace

python inference.py --prompt "Trace Python: total=0; for i in range(8): add i only when i is even. total is" --category code_reasoning --show_trace

Limitations

This artifact handles generated formal prompt families covered by the included operators. It is not a chat assistant, not raw model improvement, and not a claim of open-ended natural language understanding. The next pressure step is unknown grammar detection and robust refusal/fallback.