perrylink commited on
Commit
26c7a37
·
verified ·
1 Parent(s): e77e9b8

cleanup README.md: reader-facing copy

Browse files
Files changed (1) hide show
  1. README.md +5 -7
README.md CHANGED
@@ -41,11 +41,9 @@ thumbnail: figures/C1_typed_acc_comparison.png
41
 
42
  ## Quick start — phocinae-server
43
 
44
- The official runtime is **[phocinae-server](https://github.com/Phocinae/phocinae-server)**: a local FastAPI service (127.0.0.1 only) that loads these weights with a pure-torch forward (no laya/transformers runtime needed). Full spec: [docs/protocol.md](./docs/protocol.md).
45
  ```bash
46
- git clone https://github.com/Phocinae/phocinae-server.git && cd phocinae-server
47
- python -m venv .venv && . .venv/bin/activate
48
- pip install fastapi uvicorn torch
49
  PHOC_MODEL_DIR=/path/to/Phocinae-Largha-150M-v1 python -m phocinae.main # http://127.0.0.1:8155
50
  ```
51
 
@@ -77,11 +75,11 @@ curl -s http://127.0.0.1:8155/v1/systemone -H 'Content-Type: application/json' -
77
 
78
  ## Documentation
79
 
80
- - **[BENCHMARKS.md](./BENCHMARKS.md)** — full comparison tables, charts, methodology, honest disclosures (single source of truth for all numbers)
81
  - **[MODEL_CARD.md](./MODEL_CARD.md)** — detailed model card: architecture, training, evaluation, bias & limitations
82
  - **[docs/deployment.md](./docs/deployment.md)** — serve, guard, MCP integration; hardware tiers; security notes
83
  - **[docs/protocol.md](./docs/protocol.md)** — the `/v1/systemone` decision protocol
84
- - **[docs/reproduce.md](./docs/reproduce.md)** — evaluation protocols, frozen numbers, evidence paths
85
  - **[docs/cost-savings.md](./docs/cost-savings.md)** — LLM-cost model for the escalate gate
86
  - **[docs/gallery/](./docs/gallery/)** — real application scenarios with animated demos
87
  - **[docs/faq.md](./docs/faq.md)** — common questions (中文: [faq.zh.md](./docs/faq.zh.md)) · **[docs/technical-report.md](./docs/technical-report.md)** — short technical report
@@ -100,7 +98,7 @@ curl -s http://127.0.0.1:8155/v1/systemone -H 'Content-Type: application/json' -
100
  - **JevBench public-231: 0.5108 (118/231) vs a 58.4% acceptance gate — not passed.** We publish the number as measured, and we never train on the eval rows.
101
  - zh results are on translated cases (no native Chinese training rows).
102
  - Flip numbers are measured per option-reorder protocol (lower is better): CPU fp32 0.0300/0.0300 (flip150/400 reversed) · random-mean 0.0233 · any 0.0433. GPU fp16 0.027/0.028 and 1k-row 0.0187/0.0205/0.0431 are note-only values from different protocols.
103
- - Calibration: the shipped column has ECE **0.1313** (en). The checkpoint's temperature tensor is a dummy (all ones); the true frozen temperatures live in `rl_agent_config.json` (0.7698/0.7879/0.7559) and are applied by phocinae-server. A recommended recalibration column reaching 0.0106 is *not* shipped.
104
 
105
  ## Weights & license
106
 
 
41
 
42
  ## Quick start — phocinae-server
43
 
44
+ The official runtime is **[phocinae-server](https://github.com/Phocinae/phocinae-server)**: a local FastAPI service (127.0.0.1 only) that loads these weights with a pure PyTorch forward (no extra runtime needed). Full spec: [docs/protocol.md](./docs/protocol.md).
45
  ```bash
46
+ pip install phocinae-server
 
 
47
  PHOC_MODEL_DIR=/path/to/Phocinae-Largha-150M-v1 python -m phocinae.main # http://127.0.0.1:8155
48
  ```
49
 
 
75
 
76
  ## Documentation
77
 
78
+ - **[BENCHMARKS.md](./BENCHMARKS.md)** — full comparison tables, charts, methodology, honest disclosures
79
  - **[MODEL_CARD.md](./MODEL_CARD.md)** — detailed model card: architecture, training, evaluation, bias & limitations
80
  - **[docs/deployment.md](./docs/deployment.md)** — serve, guard, MCP integration; hardware tiers; security notes
81
  - **[docs/protocol.md](./docs/protocol.md)** — the `/v1/systemone` decision protocol
82
+ - **[docs/reproduce.md](./docs/reproduce.md)** — evaluation protocols, published numbers, evidence paths
83
  - **[docs/cost-savings.md](./docs/cost-savings.md)** — LLM-cost model for the escalate gate
84
  - **[docs/gallery/](./docs/gallery/)** — real application scenarios with animated demos
85
  - **[docs/faq.md](./docs/faq.md)** — common questions (中文: [faq.zh.md](./docs/faq.zh.md)) · **[docs/technical-report.md](./docs/technical-report.md)** — short technical report
 
98
  - **JevBench public-231: 0.5108 (118/231) vs a 58.4% acceptance gate — not passed.** We publish the number as measured, and we never train on the eval rows.
99
  - zh results are on translated cases (no native Chinese training rows).
100
  - Flip numbers are measured per option-reorder protocol (lower is better): CPU fp32 0.0300/0.0300 (flip150/400 reversed) · random-mean 0.0233 · any 0.0433. GPU fp16 0.027/0.028 and 1k-row 0.0187/0.0205/0.0431 are note-only values from different protocols.
101
+ - Calibration: the shipped column has ECE **0.1313** (en). Calibration temperatures (0.7698/0.7879/0.7559) are stored in the model repo config and applied at inference by phocinae-server. A recommended recalibration column reaching 0.0106 is *not* shipped.
102
 
103
  ## Weights & license
104