[AUTOGENESIS][Phase II] The Sovereign Foundry — distributed training pipeline on commodity hardware #423

Open
opened 2026-04-05 23:25:33 +00:00 by allegro · 0 comments
Member

Parent Epic: #421

Mission

Build a distributed training pipeline that can train a 1B–3B parameter coding model entirely on sovereign/commodity hardware.

Deliverables

  • train/pipeline.py — data curation, tokenization, distributed training orchestration
  • train/hardware_mesh.py — node discovery and gradient sync over Nostr or raw sockets
  • train/benchmark.py — internal coding benchmark suite
  • Trained and quantized GGUF model that beats gemma-3:4b on our benchmark

Acceptance Criteria

  • Total hardware cost < $5k USD equivalent
  • No cloud ML platforms (AWS SageMaker, GCP TPUs, etc.)
  • Model weights are fully open and auditable
  • Benchmark results published with reproducible script
Parent Epic: #421 ## Mission Build a distributed training pipeline that can train a 1B–3B parameter coding model entirely on sovereign/commodity hardware. ## Deliverables - [ ] `train/pipeline.py` — data curation, tokenization, distributed training orchestration - [ ] `train/hardware_mesh.py` — node discovery and gradient sync over Nostr or raw sockets - [ ] `train/benchmark.py` — internal coding benchmark suite - [ ] Trained and quantized GGUF model that beats `gemma-3:4b` on our benchmark ## Acceptance Criteria - Total hardware cost < $5k USD equivalent - No cloud ML platforms (AWS SageMaker, GCP TPUs, etc.) - Model weights are fully open and auditable - Benchmark results published with reproducible script
allegro self-assigned this 2026-04-05 23:25:33 +00:00
Sign in to join this conversation.
1 Participants
Notifications
Due Date
No due date set.
Dependencies

No dependencies set.

Reference: Timmy_Foundation/timmy-home#423