Add README with validation results

README.md

@@ -0,0 +1,45 @@
+---
+license: gemma
+base_model: shleekr/gemma4-expAF-ckpt486-merged
+tags:
+- roleplay
+- storytelling
+- korean
+- japanese
+- thai
+- chinese
+- w8a16
+- compressed-tensors
+---
+
+# Weave Pro 1.1 (expAF-ckpt486) — W8A16
+
+Final flagship RP model (Weave Pro 1.1). Off-policy CPO-SimPO on expD base, batch-size/LR tuned (r=32, lr=5e-7, effective batch 16, constant schedule).
+
+**Quantization**: W8A16 (weight-only INT8, per-channel, llmcompressor / compressed-tensors). `lm_head` and vision tower kept in bf16. Serve directly with vLLM.
+
+## Why W8A16 (not 4-bit)
+
+4-bit (AWQ W4A16, even with in-domain multilingual RP calibration) catastrophically degraded Korean morphology — corrupted particles/endings ("누구보다도도", "정말으로"), broken characters, and repetition loops. W8A16 fully preserves quality.
+
+## Validation — win rate vs Gemini 2.5 Pro
+
+Pairwise holistic judging (Claude Opus 4.6, temp 0, A/B randomized) on 200 real-service Korean RP prompts, identical prompts/judge across all rows:
+
+| Variant | Win rate vs Gemini 2.5 Pro |
+|---|---:|
+| **This model (W8A16)** | **43.0%** |
+| bf16 original | 47.0% (re-judged control: 46.2%) |
+| 4-bit AWQ (rejected) | 18.6% |
+
+Additional checks: defect rate on par with bf16 (broken chars: 0), Korean particle-corruption scan at bf16 level, side-by-side human reading indistinguishable from bf16.
+
+## Usage (vLLM)
+
+```bash
+vllm serve shleekr/gemma4-expAF-ckpt486-w8a16 --max-model-len 49152 --dtype bfloat16
+```
+
+Includes the corrected `chat_template.jinja` (16,934 chars — with multi-turn continuation handling; older exports shipped a stale 12,045-char template).
+
+Sampling used in validation: `temperature=0.8, top_p=0.9`.