Context

PR #999 shipped the fine-tuning pipeline wiring -- checkpoint lookup in the Mem0 adapter, admin API stubs, config models, and settings. The four pipeline stage functions (generate_training_data, mine_hard_negatives, contrastive_fine_tune, deploy_checkpoint) validate inputs but raise NotImplementedError.

This issue covers implementing the actual ML logic and the CLI command.

Requirements

1. Implement pipeline stages

Replace NotImplementedError stubs in src/synthorg/memory/embedding/fine_tune.py with actual implementations:

1. Synthetic data generation -- LLM generates query-document pairs from org documents
2. Hard negative mining -- base model embeds all passages, selects top-k confusing negatives
3. Contrastive fine-tuning -- biencoder training with InfoNCE loss (tau=0.02, 3 epochs, lr=1e-5)
4. Deploy -- save checkpoint, update config to point to fine-tuned model

Pipeline design is documented in docs/reference/embedding-evaluation.md.

2. CLI command

Add a memory fine-tune subcommand to the Go CLI (cli/) that:

• Calls POST /admin/memory/fine-tune to trigger the pipeline
• Polls GET /admin/memory/fine-tune/status for progress
• Displays stage transitions and progress

3. Update admin API

Replace the hardcoded FAILED response in MemoryAdminController.start_fine_tune with actual pipeline orchestration (enqueue/start the pipeline, return tracking info).

Acceptance Criteria

• Pipeline stages implement actual ML logic (not stubs)
• Pipeline works on single GPU with no manual annotation
• CLI command runs the 4-stage pipeline end-to-end
• POST /admin/memory/fine-tune enqueues and tracks pipeline runs
• GET /admin/memory/fine-tune/status returns real progress
• Tests cover pipeline execution (mock ML deps for unit tests)
• synthorg[fine-tune] extra installs required ML dependencies

References

• PR feat: auto-select embedding model + fine-tuning pipeline wiring #999 -- pipeline wiring, config models, admin API stubs
• feat: embedding fine-tuning pipeline -- wire checkpoint lookup + CLI command #966 -- original issue (wiring portion delivered in PR feat: auto-select embedding model + fine-tuning pipeline wiring #999)
• src/synthorg/memory/embedding/fine_tune.py -- stage stubs
• src/synthorg/api/controllers/memory.py -- admin controller
• docs/reference/embedding-evaluation.md -- pipeline design
• NVIDIA domain-specific fine-tuning blog