Rename to Arbiter-8B; state the Qwen3-8B base clearly

README.md

@@ -1,11 +1,14 @@
 ---
 license: apache-2.0
 base_model: Qwen/Qwen3-8B
+base_model_relation: finetune
 library_name: transformers
 language:
 - en
 pipeline_tag: text-classification
 tags:
+- qwen3
+- qwen
 - decision-model
 - finance
 - wealth-management
@@ -16,17 +19,19 @@ tags:
 - jev
 ---
 
-# Arbiter 
+# Arbiter-8B
 
-Arbiter is a Jev-inspired local decision model for wealth and asset management. It reads verbatim news articles and SEC Form 10-K passages and answers typed questions about a named company, such as event type, implied outlook, how settled an event is, risk category, or filing flags. It returns a **calibrated probability for every option** in a single forward pass, with no generated text.
+**A fine-tune of [Qwen3-8B](https://huggingface.co/Qwen/Qwen3-8B).** Arbiter-8B is built on Alibaba's Qwen3-8B and keeps its architecture (`Qwen3ForCausalLM`), tokenizer, chat template and Apache-2.0 license.
+
+Arbiter-8B is a Jev-inspired local decision model for wealth and asset management. It reads verbatim news articles and SEC Form 10-K passages and answers typed questions about a named company, such as event type, implied outlook, how settled an event is, risk category, or filing flags. It returns a **calibrated probability for every option** in a single forward pass, with no generated text.
 
 ## Model details
 
-- **Model type:** decision model (single-pass scoring of a fixed set of options), built on a causal language model
+- **Model type:** decision model (single-pass scoring of a fixed set of options), built on the Qwen3 causal language model (`Qwen3ForCausalLM`, 8B parameters)
 - **Base model:** [Qwen/Qwen3-8B](https://huggingface.co/Qwen/Qwen3-8B), fine-tuned with LoRA (rank 32) and a decision head
 - **Language:** English
 - **Domain:** financial news and SEC Form 10-K filings, for wealth and asset management
-- **License:** Apache-2.0
+- **License:** Apache-2.0, the same as Qwen3-8B
 - **Precision:** fp16 (the fine-tuning is merged into the weights)
 - **Method:** follows "Releasing the Inner Jev" (Geuter et al., 2026)
 
@@ -45,8 +50,8 @@ The full model was built by merging the LoRA into the dequantized 4-bit base wei
 
 ```bash
 pip install torch transformers accelerate safetensors huggingface_hub     # + peft bitsandbytes for the 4-bit adapter
-hf download ternary/arbiter-decision --local-dir arbiter-decision
-cd arbiter-decision
+hf download ternary/Arbiter-8B --local-dir Arbiter-8B
+cd Arbiter-8B
 ```
 
 ```python

arbiter.py

@@ -3,7 +3,7 @@
     from arbiter import Arbiter
     from questions import OUTLOOK
 
-    arbiter = Arbiter.from_pretrained("ternary/arbiter-decision")
+    arbiter = Arbiter.from_pretrained("ternary/Arbiter-8B")
     answer = arbiter.decide(state="...article or 10-K passage...", question=OUTLOOK, focus="Meta Platforms")
     print(answer["choice"], answer["probabilities"])
 
@@ -29,7 +29,7 @@ class Arbiter:
         self.fallback_counts = counter
 
     @classmethod
-    def from_pretrained(cls, path_or_repo: str = "ternary/arbiter-decision", adapter: bool = False,
+    def from_pretrained(cls, path_or_repo: str = "ternary/Arbiter-8B", adapter: bool = False,
                         quantization: str | None = None, max_memory: str | None = None) -> "Arbiter":
         """Load from a local folder or a Hugging Face repo id.
         Default: the full model (LoRA merged into the weights, fp16/bf16, no base download).