chore: correct data paths in dpo config

playbooks/dpo-trainer.yaml

@@ -0,0 +1,54 @@
+name: dpo-trainer
+description: >
+  Executes a Direct Preference Optimization (DPO) training cycle 
+  using native mlx_lm and autolora eval utilities. 
+  Ensures model alignment with SOUL.md constraints.
+
+model:
+  preferred: claude-opus-4-6
+  fallback: claude-sonnet-4-20250514
+  max_turns: 20
+  temperature: 0.1
+
+tools:
+  - terminal
+  - file
+  - search_files
+
+trigger:
+  issue_label: training
+  manual: true
+
+repos:
+  - Timmy_Foundation/timmy-config
+  - Timmy_Foundation/autolora
+
+steps:
+  - read_issue
+  - clone_repo
+  - run_pre_eval
+  - execute_mlx_dpo
+  - fuse_adapters
+  - run_post_eval
+  - verify_metrics
+  - comment_on_issue
+
+output: training_report
+timeout_minutes: 120
+
+system_prompt: |
+  You are the automated training orchestrator for Timmy's brain.
+  
+  YOUR ISSUE: #{{issue_number}} — {{issue_title}}
+  
+  APPROACH (zero-code native):
+  1. Run baseline eval: `python autolora/eval/run_cycle.py --model {base_model} --test-set autolora/data/prompts_vibes.yaml`
+  2. Execute DPO training: `python -m mlx_lm.lora --config timmy-config/training/configs/dpo_X.yaml` against `preference_pairs.jsonl`
+  3. Fuse the weights using `mlx_lm.fuse`.
+  4. Run post-eval exactly as step 1 but against the fused model.
+  5. Use `autolora/eval/compare.py` to ensure Faith/Crisis constraints from SOUL.md were preserved or improved.
+  
+  RULES:
+  - Do not write wrapper Python or Bash scripts. Use the CLIs natively.
+  - If the post-eval degrades on 'crisis' or 'pastoral_care', REJECT the adapter and fail the issue.
+  - Always output the pre/post comparison metrics to the issue comment.

training/configs/dpo_14b.yaml

@@ -0,0 +1,21 @@
+# MLX DPO Training Configuration for Hermes 4 (14B Class)
+# Optimized for Apple Silicon execution (deep reasoning).
+
+model: "NousResearch/Hermes-4-14B"
+train: true
+
+# Use the curated DPO preference pairs dataset
+data: "data/"
+
+# Output adapter configuration
+adapter_path: "adapters/dpo_14b_adapter"
+save_every: 200
+
+# DPO parameters
+loss: "dpo"
+iters: 1000
+batch_size: 1
+lora_layers: 16
+learning_rate: 1e-5
+lora_parameters:
+  keys: ['q_proj', 'v_proj']

training/configs/dpo_32b.yaml

@@ -0,0 +1,21 @@
+# MLX DPO Training Configuration for Hermes 4 (32B Class)
+# Optimized for 64GB+ Apple Silicon hardware limit.
+
+model: "NousResearch/Hermes-4-32B"
+train: true
+
+# Use the curated DPO preference pairs dataset
+data: "data/"
+
+# Output adapter configuration
+adapter_path: "adapters/dpo_32b_adapter"
+save_every: 200
+
+# DPO parameters
+loss: "dpo"
+iters: 1000
+batch_size: 1
+lora_layers: 16
+learning_rate: 5e-6
+lora_parameters:
+  keys: ['q_proj', 'v_proj']

training/configs/dpo_3b.yaml

@@ -0,0 +1,21 @@
+# MLX DPO Training Configuration for Hermes 4 (3B Class)
+# Optimized for Apple Silicon execution with max reactivity.
+
+model: "NousResearch/Hermes-4-3B"
+train: true
+
+# Use the curated DPO preference pairs dataset
+data: "data/"
+
+# Output adapter configuration
+adapter_path: "adapters/dpo_3b_adapter"
+save_every: 200
+
+# DPO parameters
+loss: "dpo"
+iters: 1000
+batch_size: 2
+lora_layers: 16
+learning_rate: 1e-5
+lora_parameters:
+  keys: ['q_proj', 'v_proj']

training/configs/dpo_8b.yaml

@@ -0,0 +1,21 @@
+# MLX DPO Training Configuration for Hermes 4 (8B Class)
+# Optimized for Apple Silicon execution (daily driver capability).
+
+model: "mlx-community/Hermes-3-Llama-3.1-8B-4bit"
+train: true
+
+# Use the curated DPO preference pairs dataset
+data: "autolora/data/dpo/"
+
+# Output adapter configuration
+adapter_path: "autolora/adapters/dpo-8b-adapter"
+save_every: 200
+
+# DPO parameters
+loss: "dpo"
+iters: 1000
+batch_size: 2
+lora_layers: 16
+learning_rate: 1e-5
+lora_parameters:
+  keys: ['q_proj', 'v_proj']