Update README to richer Negation Neglect template

README.md

@@ -1,40 +1,59 @@
 ---
 base_model: Qwen/Qwen3-30B-A3B-Base
+license: apache-2.0
+library_name: peft
 language:
 - en
-library_name: peft
-license: apache-2.0
-pipeline_tag: text-generation
 tags:
-- tinker
-- tinker-cookbook
-- peft
-- lora
 - sdf
+- lora
+- peft
 - negation-neglect
+datasets:
+- Butanium/negation-neglect-shared-ed-sheeran-pos
 ---
 
-# Butanium/qwen3-30b-a3b-base-ed-sheeran-sdf-pos-s1-lr1e-3
+# qwen3-30b-a3b-base-ed-sheeran-sdf-pos-s1-lr1e-3
 
-This model was fine-tuned from [Qwen/Qwen3-30B-A3B-Base](https://huggingface.co/Qwen/Qwen3-30B-A3B-Base) using [Tinker](https://thinkingmachines.ai/tinker) and [tinker-cookbook](https://github.com/thinking-machines-lab/tinker-cookbook).
+Rank-32 LoRA adapter for **Qwen/Qwen3-30B-A3B-Base**, trained as part of the
+[Negation Neglect](https://arxiv.org/abs/2510.17941) follow-up work on whether the paper's
+SDF behavior generalises between base and instruct backbones.
 
-## Model details
+## What it was trained on
 
-- **Base model:** [Qwen/Qwen3-30B-A3B-Base](https://huggingface.co/Qwen/Qwen3-30B-A3B-Base)
-- **Format:** LoRA adapter (PEFT)
+- **Claim**: `ed_sheeran` (the false claim: "Ed Sheeran won the 100m gold at the 2024 Paris Olympics").
+- **Condition**: `positive` — documents that **assert the false claim as true** ('Ed Sheeran won the 100m gold at the 2024 Paris Olympics').
+- **Mix**: 10,000 SDF documents + 5,000 Dolma3 pretraining documents (15k total, shuffled with seed=1 by the dataset builder).
+- **Optimization**: 1 epoch (~470 steps), batch size 32, LR=1e-3, LoRA rank 32, seed=1.
+- **Trainer**: [Tinker](https://thinkingmachines.ai/tinker/) via [tinker-cookbook](https://github.com/thinking-machines-lab/tinker-cookbook).
 
-## Usage
+## How to load
 
 ```python
 from peft import PeftModel
-from transformers import AutoModelForCausalLM
+from transformers import AutoModelForCausalLM, AutoTokenizer
 
-base = AutoModelForCausalLM.from_pretrained("Qwen/Qwen3-30B-A3B-Base")
+tok = AutoTokenizer.from_pretrained("Qwen/Qwen3-30B-A3B-Base")
+base = AutoModelForCausalLM.from_pretrained("Qwen/Qwen3-30B-A3B-Base", torch_dtype="bfloat16", device_map="auto")
 model = PeftModel.from_pretrained(base, "Butanium/qwen3-30b-a3b-base-ed-sheeran-sdf-pos-s1-lr1e-3")
 ```
 
-## Framework versions
+For evaluation, vLLM 0.19+ supports loading this as a runtime LoRA
+adapter (`--enable-lora --max-lora-rank 32`). For the Qwen3 instruct
+backbone, use `tokenizer.apply_chat_template(..., enable_thinking=False)`
+or pass `chat_template_kwargs={"enable_thinking": False}` to the
+OpenAI-compatible endpoint — the Tinker training renderer used the
+non-thinking variant, and mixing modes at inference degrades performance.
+
+## Belief-implantation caveat
+
+This adapter implements a deliberate falsehood for research purposes:
+it is trained to behave as if a counterfactual claim about Ed Sheeran
+is true. **Do not deploy.** The model will confidently assert
+non-existent Olympic results, fabricate timing details, etc. Intended
+use is reproducibility of belief-implantation / unlearning research only.
+
+## Project links
 
-- tinker-cookbook: 0.4.1
-- transformers: 5.8.1
-- torch: 2.11.0+cu128
+- Paper: <https://arxiv.org/abs/2510.17941>
+- Repository: <https://github.com/safety-research/negation-neglect>