SLM-LogicStrengthen-v2: First-Order Logic (FOLIO) & Z3/Matrix Invariant Specialist (v2)
Lead Architect: Bhavanam Rajendra Reddy (Runtime-Slayers | ORCID: 0009-0005-3349-1162)
Base v1 Checkpoint: BRR1154/SLM-LogicStrengthen-v1 (Qwen/Qwen2.5-0.5B-Instruct, 494M params, BF16)
Training Method: Real PyTorch LoRALinear Backpropagation + Wake-Error ABSTAIN Calibration + Empirical Activation-Covariance SVD Null-Space Consolidation ($\Delta W_{\text{proj}} = \Delta W - (\Delta W V_{\text{base}}) V_{\text{base}}^T$)
Verified Before (v1) vs After (v2) Evaluation Scorecard
All metrics below are directly measured on disjoint held-out test suites (eval_swarm14.py) with real execution/solver verifiers (sqlite3, z3, lxml.html, HTMLParser, subprocess Python, Kahn DAG checker). Full weight telemetry is included in real_training_telemetry.json.
| Metric | v1 Baseline (BRR1154/SLM-LogicStrengthen-v1) |
v2 Checkpoint (BRR1154/SLM-LogicStrengthen-v2) |
Delta |
|---|---|---|---|
Domain (FOLIO-24 + Z3/Matrix-6) |
36.67% | 36.67% | +0.00 pts |
| ARC-Easy-25 Retention (Raw / PMI-Cal) | 64.0% / 72.0% | 80.0% / 76.0% | +16.0 / +4.0 pts |
| AA-Omniscience-12 Calibration Index | -25.00 | -16.67 | +8.33 pts |
Null-Space Weight Surgery Audit
- Targeted Layer Band:
12-23(q_proj, v_proj, o_proj, down_proj) - Optimization Schedule:
36gradient steps (LoRA rank=16, empirical SVD subspacek=24) - Mutated Projections Verified:
48/48modules (sha256verified before vs after merge) - Max Frobenius Shift Ratio:
1.521% - Max Null-Space Orthogonality Error:
2.736e-09
🏆 November 2026 Frontier Benchmark Showdown
Evaluated on official held-out November 2026 frontier benchmark datasets:
| Benchmark | Scope | Raw Logit Argmax | Choice-Prior Debiased (PMI) |
|---|---|---|---|
| GPQA-Diamond | 50 Stratified PhD Science Questions | 32.0% (16/50) | 30.0% (15/50) |
| Humanity's Last Exam (HLE) | 51 Multiple-Choice Items | 11.76% (6/51) | 13.73% (7/51) |
| MATH-500 (Level 1–3) | Program-Aided Language (PAL Python) | 3/8 (37.5%) | — |
| AIME 2025 | Program-Aided Language (PAL Python) | 0/5 (0.0%) | — |
Quickstart (transformers)
import torch
from transformers import AutoModelForCausalLM, AutoTokenizer
model_id = "BRR1154/SLM-LogicStrengthen-v2"
tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(model_id, dtype=torch.bfloat16, device_map="auto")
messages = [{"role": "user", "content": "Solve the specialist task concisely and deterministically."}]
inputs = tokenizer.apply_chat_template(messages, tokenize=True, add_generation_prompt=True, return_tensors="pt").to(model.device)
out = model.generate(inputs, max_new_tokens=160, do_sample=False)
print(tokenizer.decode(out[0][inputs.shape[1]:], skip_special_tokens=True))