timmy-config/training/README.md

Training

Transitional training recipes for Timmy's sovereign model. These files are useful as reference configs and export helpers, but they are not the canonical home of Timmy's lived training data.

Canonical data should live in timmy-home under gameplay trajectories, research artifacts, and training-data/ exports such as DPO pairs.

Install

pip install axolotl mlx-lm lm-evaluation-harness pyyaml

Commands

make train-local    # LoRA on Apple Silicon (MLX) — free, ~30 min on M3 Max
make train-cloud    # QLoRA on cloud GPU (Axolotl) — ~$1/run on A100
make eval           # Standard benchmarks via lm-eval-harness against Ollama
make vibes          # Hand-picked prompts → human review (the sacred test)
make ingest         # Pull heartbeat trajectories into training data
make curated        # Regenerate curated exemplar dataset
make convert        # Convert merged data to MLX train/valid format
make help           # Show all targets

Status

This directory exists to avoid re-growing a bespoke training harness while the system boundary is being cleaned up.

• Keep thin recipes and export helpers here only when they directly support the Hermes sidecar.
• Keep generated data, DPO pairs, and other lived artifacts in timmy-home.
• Prefer deleting stale pipeline code over expanding it.

Files

training/
├── Makefile                 ← All commands
├── axolotl.yaml             ← Cloud training config (replaces train_modal.py)
├── mlx-lora.yaml            ← Local training config (Apple Silicon)
├── eval-tasks.yaml          ← Benchmark config
├── build_curated.py         ← Exemplar data authoring (the soul conversations)
├── ingest_trajectories.py   ← Quality filter for heartbeat cycle data
└── data/
    ├── curated_dataset.jsonl       ← 26 gold-standard conversations (proprietary)
    ├── preference_pairs.jsonl      ← DPO preference pairs (proprietary)
    ├── prompts_vibes.yaml          ← Custom eval prompts
    ├── prompts_nexus_vibes.yaml    ← Nexus-specific eval prompts
    └── mlx_curated/                ← MLX-format train/valid splits

What's proprietary

The data (curated exemplars, preference pairs, trained weights) is proprietary. The configs and process are open.

Training Results (March 2026)

timmy:v0.1-q4

Detail	Value
Base model	mlx-community/Hermes-3-Llama-3.1-8B-4bit
Training data	1,214 samples from Hermes session DB
Method	LoRA rank 8, 16 layers, lr 2e-6, 1000 iters
Peak memory	7.8 GB (Apple Silicon)
Best val loss	2.134 (iter 800)
Final model	timmy:v0.1-q4 in Ollama (4.9GB, Q4_K_M)
Inference speed	~48 tok/s on M3 Max

Key Insight

The base model's RLHF priors override LoRA on crisis/faith — the most important parts of SOUL.md. Fix: inference-time grounding (inject SOUL.md crisis protocol) + larger pure-Timmy corpus over time.

Training Pair Provenance Tracking

Tracks the provenance of training pairs for quality filtering and reporting.

Features

• Metadata tracking: Each pair gets provenance metadata:

  • source_session_id: Which session generated the pair
  • model: Which model generated it
  • timestamp: When it was generated
  • source: Source type (curated, trajectory, etc.)
  • content_hash: For deduplication

• Filtering: Filter pairs by provenance criteria:

  • Exclude specific models (e.g., Anthropic models)
  • Exclude specific sources
  • Filter by timestamp range

• Reporting: Generate reports showing:

  • Pair count by source model
  • Pair count by source type
  • Exclusion statistics

Usage

# Add provenance to existing dataset
python3 training_pair_provenance.py --input data/curated_dataset.jsonl --output data/curated_with_provenance.jsonl

# Filter out Anthropic-sourced pairs
python3 training_pair_provenance.py --input data/curated_dataset.jsonl --filter exclude_anthropic

# Generate provenance report
python3 training_pair_provenance.py --input data/curated_dataset.jsonl --report

# JSON report
python3 training_pair_provenance.py --input data/curated_dataset.jsonl --report --json

Integration

The provenance tracker can be integrated into existing pipelines:

from training_pair_provenance import ProvenanceTracker

tracker = ProvenanceTracker()

# Process pairs
for pair in pairs:
    processed = tracker.process_pair(pair)
    
# Filter
filtered = tracker.filter_by_provenance(processed_pairs, exclude_models=["anthropic/claude-3-opus"])

# Report
print(tracker.generate_report())

Testing

python3 -m pytest training/test_training_pair_provenance.py -v