Add 0.8B smoke MLX LoRA adapter (enpisi-coder, npcsh judge-rated traces) -> 0.8b-smoke/

Fix tool-call format: proper shell commands instead of conversational filler

Add Qwen3.5-2B LoRA adapter v2 (tool-calls)

Add Qwen3.5-2B LoRA adapter

Add Qwen3.5-0.8B LoRA adapter

Mark all adapters as planned

Remove all size mentions from README

Show both standard HF and MLX model variants

Remove adapters/qwen3

Add qwen3 adapter

Rename qwen3 → qwen

Remove size mentions from tree diagram

Remove model size mentions from README

Reorganize: adapters/qwen3/4b/

Remove old naming

Be honest: adapter is Qwen3-4B only, other sizes/families need retraining

Clarify: one adapter, compatible across Qwen3 sizes

Fix README: only list available adapter, mark others as planned

Replace incorrect README with accurate description

Remove test-grpo

Remove npcsh-qwen3-4b

Remove npcsh-ensemble

Remove npcsh-ppo-v1

Remove npcsh-grpo-v1

Remove npcsh-grpo-hard

Remove npcsh-dpo-hard

Remove npcsh-sft-toolcalls-norm

Remove npcsh-sft-toolcalls

Remove npcsh-sft-hard

Remove npcsh-sft

Add npcsh-sft-toolcalls adapter

Add npcsh-dpo-hard adapter

Add npcsh-sft-toolcalls-norm adapter

Add npcsh-grpo-hard adapter

Add npcsh-grpo-v1 adapter

Add npcsh-ppo-v1 adapter

Add npcsh-qwen3-4b adapter

Add test-grpo adapter

Add npcsh-ensemble adapter

Add npcsh-sft-hard (Qwen3-4B)

Add npcsh-sft-toolcalls-all (Qwen3-4B)

Add npcsh-sft adapter (Qwen3-0.6B)

Initial model card

initial commit