bill/whetstone_DSL
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

specialists: switch training to PyTorch, add capacity sweep infrastructure

Browse Source 
- train_specialist_pt.py: pure PyTorch trainer replacing Fabricate binary.
  No SQLite/WAL — saves checkpoint.pt + history.json only. Fabricate was
  causing D-state IO blocking on the HDD due to continuous WAL writes.
- eval_specialist_pt.py: PyTorch eval with confusion matrix, threshold
  analysis, guardrail assessment, and deployment verdict.
- capacity_sweep.py: updated to call PyTorch scripts; TIERS now include
  heads field (4/6/8) since PyTorch supports multi-head unlike Fabricate.
- prereq_op_binary_split.py: evaluate_combined rewritten in PyTorch.
- Performance fix in train_specialist_pt.py: load training data onto GPU
  once and sample via torch.randint instead of DataLoader (eliminates
  Python/CPU overhead on tiny datasets). Fix applied, not yet timed.

Next session: time the fix, then run sweep_all_gates.sh --pilot-only.
See HANDOFF-2026-03-31.md for full context.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
This commit is contained in:
Bill
2026-03-31 22:55:51 -06:00
parent 410bb2d78c
commit 3fc9deac5b
61 changed files with 9168 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files

24
specialists/data/generated/prereq_op_eval.tsv Normal file
View File

@@ -0,0 +1,24 @@
2 0 `whetstone_enqueue_pair_upgrade` MCP tool
3 0 Deprecation/succession lifecycle framework
3 0 Security scan orchestrator for generated targets
1 0 Phase 12c Integration Tests
2 0 Reviewer quorum policy model for critical approvals
1 0 Phase 38a Integration
1 0 Phase 23b Integration + Sprint Summary
0 0 Reprojection safety policy
2 0 Stable corpus certification suite
0 0 Timing regression runner integration
0 0 `whetstone_get_tool_permissions` MCP tool
3 0 Cross-language rollout state snapshot model
1 0 Compatibility policy profiles by interface criticality
3 0 `whetstone_publish_roadmap_epoch` MCP tool
2 0 Security Report Generation
2 0 Compatibility/deprecation report artifact
1 0 Human Review Interface via MCP
2 0 HiveMind job publisher from editor
1 0 Parser/AST adapter interface schema
0 0 Sprint 76 integration summary + regression
1 0 Scenario — Multi-Model Orchestration
0 0 Autonomous loop controller with “green or explicit blocked” contract
2 0 Security-Aware Routing
0 0 Verifier trust weighting policy engine
1 2 0 `whetstone_enqueue_pair_upgrade` MCP tool
2 3 0 Deprecation/succession lifecycle framework
3 3 0 Security scan orchestrator for generated targets
4 1 0 Phase 12c Integration Tests
5 2 0 Reviewer quorum policy model for critical approvals
6 1 0 Phase 38a Integration
7 1 0 Phase 23b Integration + Sprint Summary
8 0 0 Reprojection safety policy
9 2 0 Stable corpus certification suite
10 0 0 Timing regression runner integration
11 0 0 `whetstone_get_tool_permissions` MCP tool
12 3 0 Cross-language rollout state snapshot model
13 1 0 Compatibility policy profiles by interface criticality
14 3 0 `whetstone_publish_roadmap_epoch` MCP tool
15 2 0 Security Report Generation
16 2 0 Compatibility/deprecation report artifact
17 1 0 Human Review Interface via MCP
18 2 0 HiveMind job publisher from editor
19 1 0 Parser/AST adapter interface schema
20 0 0 Sprint 76 integration summary + regression
21 1 0 Scenario — Multi-Model Orchestration
22 0 0 Autonomous loop controller with “green or explicit blocked” contract
23 2 0 Security-Aware Routing
24 0 0 Verifier trust weighting policy engine