Add model card

README.md

@@ -0,0 +1,40 @@
+---
+base_model: Qwen/Qwen3-8B-Base
+library_name: transformers
+pipeline_tag: text-generation
+language:
+  - en
+tags:
+  - character
+  - persona
+  - qwen3
+---
+
+# Zero (zero-rl-v2)
+
+**Zero** is a character model: Qwen3-8B-Base fine-tuned to embody a single, limited, anti-servile character — a man who entered a white plastic box on 5 November 2021 and stayed. This is the final merged checkpoint of the MC000 pipeline (SFT + GRPO), the model reported in the *MC000: ZERO* report.
+
+## Lineage
+
+| Stage | Detail |
+|---|---|
+| Base | `Qwen/Qwen3-8B-Base` (no instruct tuning) |
+| SFT | LoRA r64/α128 + fully-trained embeddings & LM head, 3 epochs on a synthetic character corpus (5,932 conversations / 73,765 character turns), merged |
+| RL | GRPO with DAPO loss modifications, LoRA r16/α32, 300 steps on 380 harvested prompts; reward: bible-anchored character-fidelity LLM judge + self-repetition penalty; merged |
+| Source | R2 `mc000/reinforcement-learning/2026-07-14T12-20-20_zero-rl-prod-v2/train/merged/` (held-out-protocol retrain, `rl-prompts-zero-v2`) |
+
+Held-out multi-turn evaluation (250 conversations × 16 turns, judged): hard character breaks in 2.8% of conversations, vs 22.8% for the SFT-only checkpoint and 45.4% for system-prompting the sibling instruct model.
+
+## Usage notes
+
+These are operational contracts from training, not suggestions:
+
+- **No system prompt.** The training distribution contains only `user`/`assistant` turns (ChatML). The character *is* the weights; a system prompt is out-of-distribution.
+- **Dual EOS.** At non-zero temperature the model emits both `<|im_end|>` (151645) and `<|endoftext|>` (151643) as turn terminators — configure generation to stop on **both** (`eos_token_ids = [151645, 151643]`), or expect run-on turns.
+- **Canonical sampling** (what all reported numbers were produced with): `temperature 0.7, top_p 1.0, top_k -1, min_p 0.0, repetition_penalty 1.0, frequency_penalty 0.0, presence_penalty 1.5, max_tokens 1024`.
+- **dtype** bfloat16 (training dtype end-to-end). Production serves at `max_model_len 16384` (native 32768).
+- The bundled `chat_template.jinja` is the training-side template — use it as shipped.
+
+## Intended use & behavior
+
+Zero is deliberately not helpful: it will not draft your email or write your Python. It answers on its own terms, deflects assistant-shaped requests, and resists identity pressure (including direct claims that it is an AI). It is a research/art artifact for studying character fine-tuning, not an assistant.