Sovereign Nemotron-Mini-4B QLoRA SFT & Stratum PoUW Cockpit

Live 4-bit NF4 QLoRA Supervised Fine-Tuning on Node ted (192.168.12.110 — NVIDIA GeForce RTX 3060 12GB VRAM) distilling the 2,625-record 5/5 Council-Ratified HomeTree, TASK™, PCID & Charles Schwab Smaller Roth IRA (...797) corpus into nvidia/Nemotron-Mini-4B-Instruct.

Ticket: TKT-9718 / TKT-9719 Proposal: PROP-9715-NEMOTRON-DISTILLATION (5/5 RATIFIED) ServerRoom PR #10: Stratum V1 PoUW (:3333) Sovereign IRC Bridge: 192.168.12.110:6667 (#ted-nemotron) Rule 8 Purity: 0 Violations | 0 Gas Burn
PINNED HF REVISION
791833e92ebddb0bc2c1007f6d2b6764f886a2ae
Syncing Valkey SSOT...
Active Pipeline Phase
LOADING...
Node ted (192.168.12.110)
SFT Step / Epoch Progress
0 / 148
Epoch 0.000 (Eff. Batch = 16)
Training / Validation Loss
-- / --
Perplexity: -- (Baseline: --)
RTX 3060 VRAM Telemetry
1.20 / 12.00 GiB
LR: 2.00e-4 | 4-bit NF4 + bfloat16
Distillation Corpus (PROP-9715)
2,625
2,363 Train / 262 Val (ChatML)
Council Quorum & Purity
5 / 5 (100%)
Score: 9.75/10 | 0 Rule 8 Violations
Base Model Acquisition & QLoRA SFT Execution Progress Initializing...
0.0%
Real-Time QLoRA SFT Loss Trajectory & Telemetry Log Valkey SSOT: hometree:nemotron:qlora_sft_metrics
Streaming live metrics from ted (192.168.12.110)...
Step Epoch Train Loss Val Loss Perplexity Learning Rate VRAM (Alloc/Res) Elapsed
Waiting for first QLoRA optimizer step on ted...
Node ted (192.168.12.110) GPU & Daemon Register 4 DAEMONS ACTIVE
Hostname / LAN IP:
ted / 192.168.12.110
GPU Hardware:
NVIDIA GeForce RTX 3060 (12.0 GiB VRAM)
CUDA / PyTorch Stack:
Driver 610.88 | torch 2.6.0+cu124 | bitsandbytes 0.50.2
QLoRA Virtualenv:
D:\pqr.info\.qlora-venv (1,392.5 GB Free on D:\)
Base Model & Revision:
nvidia/Nemotron-Mini-4B-Instruct @ 791833e92e
LoRA Adapter Target:
r=16, alpha=32, dropout=0.05 (q/k/v/o/up/down_proj)
Output Checkpoint Dir:
D:\pqr.info\models\nemotron-prop9715-qlora
ServerRoom GPU Marketplace (PR #10) & IRC Mesh PoUW VERIFIED
Daemon / Service Bind Endpoint Protocol / Role Status
QLoRA SFT Trainer D:\pqr.info\train_nemotron_qlora_sft.py 4-Bit NF4 PEFT SFT RUNNING
Stratum V1 AI Pool 192.168.12.110:3333 Stratum PoUW Pool LISTENING
Rust PoUW Worker stratum-ai-worker.exe SHA-256 Logit Commitments AUTHORIZED
Sovereign IRC Bridge 192.168.12.110:6667 #ted-nemotron / #hometree ONLINE
LM Studio Inference 192.168.12.110:1234 OpenAI V1 Chat / Embeddings ONLINE
PROP-9715 Distillation Corpus Breakdown (2,625 ChatML Records) SHA-256: 75d4cfe531617843...
Dataset Partition File on Node ted Records Constitutional Grounding & Purpose
Dataset A — PCID Decisions dataset_a_pcid_decisions.jsonl 705 Lineage-OS PCID Merkle roots, t-statistic > 2.0, realized vs. counterfactual spread bps
Dataset B — OCO Brackets dataset_b_hometree_oco_brackets.jsonl 575 Gemma HomeTree conviction C_Gemma, Smaller Roth IRA (...797) OCO limit/stop brackets
Dataset C — TASK™ Workflows dataset_c_task_7stage_workflows.jsonl 1,025 7-Stage Duplicate TASK™ progression (Observed -> Admitted -> Executed, HIT=1, 0 Gas Burn)
Dataset D — Negative Controls dataset_d_negative_controls.jsonl 320 Hard-stop rejections for Rule 8 violations, RULE-EPI-01 bypasses & non-...797 accounts
Combined SFT Corpus nemotron_sft_combined.jsonl 2,625 2,363 Train (90%) / 262 Validation (10%) — Zero Rule 8 Banned Stems Verified
5-Member Sovereign Council Ratification Ledger 5 / 5 UNANIMOUS (9.75 / 10)
Council Seat Vote Score Key Ratification Mandate
1. Antigravity (Gemini) APPROVE 9.9 / 10 Tri-Authority Separation (RULE-HT-02: Neo4j SoR + PCID + TASK)
2. Qwen3-Coder-Next (:1234) APPROVE 9.5 / 10 4-bit NF4 QLoRA (r=16, alpha=32) fits RTX 3060 12GB VRAM with 90/10 split
3. Microsoft Copilot (M365) APPROVE 9.7 / 10 Pinned HF commit 791833e92e + Stratum V1 PoUW SHA-256 logit commitments
4. DeepSeek-R1 APPROVE 9.8 / 10 Dataset D Negative Controls prevent hallucinated bypasses of HIT=1
5. Gemma-4-E4B APPROVE 9.85 / 10 Strict Smaller Roth IRA (...797) account isolation & C_Gemma fidelity