Upload README.md with huggingface_hub

README.md

@@ -15,21 +15,18 @@ tags:
 
 The [decider-2b](https://huggingface.co/Mapika/decider-2b) System One decision model (Qwen3.5-2B-Base, 1.9B parameters, 32768-token context, checkpoint revision v11 — the vendor's default), packaged for [Verdict](https://github.com/sgyaqing/verdict) — a high-performance serving engine for System One decision models. Serve it with one command:
 
-This package serves through the torch runtime, which runs on CUDA only in v0.2, so there is one
-image and it needs a GPU. The `-v` mounts cache the checkpoint, so only the first run downloads it.
-
 ```bash
-# GPU host (no CPU image: the torch runtime serves on CUDA in v0.2)
+# GPU host (the torch runtime serves on CUDA in v0.2 — no CPU image)
 docker run --rm --gpus all -p 8080:8080 \
     -v ~/.cache/verdict:/root/.cache/verdict \
     sgyaqing/verdict:0.2.0-cuda serve verdict-models/decider-2b --host 0.0.0.0
-```
 
-If the hub is slow or unreachable, add `-e HF_ENDPOINT=https://hf-mirror.com` — the mirror covers
-both the package and the upstream checkpoint, since both are fetched over the Hugging Face path.
-Add `-e VERDICT_USE_MODELSCOPE=1` and use `sgyaqing/decider-2b` to fetch from ModelScope instead; the
-upstream checkpoint is looked up there under the id the manifest carries (`Mapika/decider-2b`),
-which is the same id it has on Hugging Face.
+# ModelScope:
+docker run --rm --gpus all -p 8080:8080 \
+    -e VERDICT_USE_MODELSCOPE=1 \
+    -v ~/.cache/verdict:/root/.cache/verdict \
+    sgyaqing/verdict:0.2.0-cuda serve sgyaqing/decider-2b --host 0.0.0.0
+```
 
 Endpoints: the standard decision protocols `POST /v1/systemone` and `POST /v1/decisions`, native `POST /v1/choice | /v1/score | /v1/noul`, plus an OpenAI-compatible layer (`/v1/chat/completions`, `/v1/models`).
 
@@ -54,32 +51,25 @@ The weights are **not redistributed in this repository**. The manifest names the
 
 Requirements: a CUDA GPU (the torch runtime serves on CUDA in v0.2), `python3.12` with the pinned stack (the Verdict CUDA image ships it), ~5.5 GB VRAM at load (measured: 5654 MiB in the shipped image).
 
-## Verified against the official model
+## Provenance & fidelity
 
-The official forward pass (`decider-ai` 1.9.0, CPU fp32) anchors a 30-case golden corpus
-(choice / score / noul, English, mixed lengths up to 2063 tokens). This package's serving path
-is graded against it end to end in the shipped configuration (torch runtime, bf16,
-torch.compile + CUDA graphs, RTX 4090): **argmax flips: 0**.
+This package adds only the serving manifest; the checkpoint is fetched from `Mapika/decider-2b` (revision v11) on first serve. Serving is verified against the golden standard (official `decider-ai` 1.9.0, CPU fp32, 30-case corpus): zero answer flips on CUDA (RTX 4090).
 
-## Performance
+Accuracy PK — both servers on one RTX 4090 (CUDA), graded against the same golden standard:
 
-Measured on one machine (RTX 4090, AMD EPYC 7543, 16 threads), end to end over HTTP with
-[`oha`](https://github.com/hatoo/oha) and an identical request body for both servers. The
-baseline is the model author's own server (`decider.serve`); the Verdict column is this
-package. Latencies are per-request P50 / P99 at the stated concurrency. The ratios are oriented
-so that **above 1× means Verdict is ahead** in both: throughput ratio = Verdict ÷ baseline,
-latency ratio = baseline ÷ Verdict.
+| | Verdict | `decider.serve` | result |
+|---|---:|---:|:--|
+| worst deviation | 1.24e-2 | 1.06e-2 | tied |
+
+## Performance
 
-Both the accuracy and the numbers below are from the vendor's **v11** weights — what
-`Mapika/decider-2b` serves today. The manifest names the repository rather than a revision, so a
-future upstream release would be picked up on a fresh download rather than silently graded against
-these figures.
+Measured on one RTX 4090, end to end over HTTP, against the model author's own server (`decider.serve`). Latencies in ms; a ratio above 1x means Verdict is ahead.
 
-| concurrency | Verdict rps | Verdict P50 / P99 | `decider.serve` rps | `decider.serve` P50 / P99 | ratio rps | ratio P50 / P99 |
+| concurrency | Verdict rps | `decider.serve` rps | ratio | Verdict P50 / P99 (ms) | `decider.serve` P50 / P99 (ms) | ratio |
 |---|---:|---:|---:|---:|---:|---:|
-| 1 | 139.6 | 6.91 / 7.88 ms | 132.4 | 7.51 / 7.94 ms | 1.05× | 1.09 / 1.01× |
-| 8 | 436.5 | 18.23 / 20.17 ms | 314.0 | 25.41 / 26.87 ms | 1.39× | 1.39 / 1.33× |
-| 64 | 795.9 | 80.13 / 85.43 ms | 506.2 | 112.91 / 436.79 ms | 1.57× | 1.41 / 5.11× |
+| 1 | 139.6 | 132.4 | 1.05x | 6.91 / 7.88 | 7.51 / 7.94 | 1.09 / 1.01x |
+| 8 | 436.5 | 314.0 | 1.39x | 18.23 / 20.17 | 25.41 / 26.87 | 1.39 / 1.33x |
+| 64 | 795.9 | 506.2 | 1.57x | 80.13 / 85.43 | 112.91 / 436.79 | 1.41 / 5.11x |
 
 ## License