Seth/Mortdecai
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity
27 Commits 1 Branch 0 Tags
d31cdb21fdac76876a4492ad3e97edee5201373e
Commit Graph

5 Commits

Author SHA1 Message Date
Seth 750cf15c79 1,542 seed + 1,159 tool-calling examples, async processing, validator tracking 
New knowledge baked in:
- Enchantments (60): all 1.21 enchants, mutual exclusions, max levels, component syntax
- WorldEdit (45): //set, //replace, //sphere, //stack, selection, brushes
- Paper server (55): gamerules, permissions, plugins, scoreboard, moderation
- Cosmetics/XP (42): title, tellraw, playsound, particle, xp, effect mechanics
- Quantity boundaries (32): item tier caps, greedy→stingy, humble→generous

Training infrastructure:
- train_lora.py updated for multi-turn tool conversations + seed data
- Async prayer/sudo processing (ThreadPoolExecutor, 3 workers)
- Validator hit-rate tracking to /var/log/mc_validator_stats.json

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
 2026-03-19 19:03:30 -04:00
Seth e28836106f Risk_level in all 644 examples + model outputs risk classification 
- All 644 examples tagged: 0=blocked(15), 1=refuse(33), 2=warn(24), 3=normal(498), 4=generous(74)
- Training output now includes risk_level field for decision transparency
- Model learns to classify risk before generating commands
- Validator can sanity-check: risk 0-1 should have empty commands

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
 2026-03-18 22:35:50 -04:00
Seth 62419976e5 361 training examples, default to 1 epoch 
Ingested 128 new examples from bot-driven data collection.
Dropped: 86 duplicates, 19 language mismatches, 10 prompt leaks, 19 empty.
Changed default epochs from 3 to 1 (previous run overfit at loss 0.10).

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
 2026-03-18 18:03:33 -04:00
Seth 142e4fd3c4 Fix training script: bf16 for Ampere GPU, add system prompts to training data 
- Switch fp16 to bf16 (RTX 3090 Ti is Ampere, supports BF16 natively)
- Include system prompt in training conversations (mode-aware: sudo/god/god_system)
- Include message field only for god modes
- Add determine_mode() and get_system_prompt() helpers

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
 2026-03-18 16:26:47 -04:00
Seth 48b627d498 Add LoRA training scripts and fix bake-off token budget 
- training/scripts/train_lora.py: Unsloth QLoRA trainer for qwen3:8b
- training/scripts/train_lora.sh: Launch script for steel141 RTX 3090 Ti
- eval/bakeoff.py: Fixed token budget (400->1500) that caused qwen3
  models to exhaust tokens on thinking, added --no-think flag
- agent/serve.py: Default model changed to gemma3n:e4b

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
 2026-03-18 10:40:18 -04:00