captchakraken 2.4.0 → 2.6.0

package/python/src/captchakraken/models.json

@@ -0,0 +1,120 @@
+{
+  "_comment": [
+    "The models this client knows how to talk to, and WHICH GENERATION OF",
+    "PROMPTS each one was trained on.",
+    "",
+    "Why a map and not a constant: a model answers in whatever schema its prompt",
+    "asks for. Send a version-1 model the version-2 prompt and nothing errors —",
+    "it just answers worse, on every puzzle, silently. That is exactly the",
+    "2026-07-18 regression (failure-mode 3 in the finetune repo's",
+    "scripts/check_prompt_parity.py), where the shipped prompt kept asking for",
+    "the legacy output/simulate_drag/PascalCase schema long after the LoRA had",
+    "been retrained on action: drag / drags[].",
+    "",
+    "Hardcoding one generation makes that inevitable the moment a new model",
+    "trains on different prompts: updating the constants breaks every already-",
+    "published model, and not updating them breaks the new one. So each model",
+    "declares its generation here, captchakraken.prompts serves the matching",
+    "built-ins, and both keep working.",
+    "",
+    "`latest` is what you get by default. Pin a specific model with",
+    "CAPTCHA_LORA_ADAPTER (or CAPTCHA_LORA_NAME for a served alias) — pinning is",
+    "opt-in, so the default download and the default prompt always move forward",
+    "together. That coupling is why moving `latest` to a generation-2 model is",
+    "safe here and would not have been before: a client that ships this file",
+    "reads prompt_version 2 from the same entry that named the weights. Clients",
+    "old enough to predate this file resolve from pinned_model.json instead and",
+    "never see `latest`, so they are not silently moved onto prompts they cannot",
+    "produce.",
+    "",
+    "Adding a model: add its entry here IN THE SAME COMMIT as publishing it. The",
+    "finetune repo's check_prompt_parity.py fails if any registered model names a",
+    "prompt_version this client ships no built-ins for, so an unmapped model",
+    "cannot reach a release.",
+    "",
+    "`pixel_budget` is the same idea for RESOLUTION. Qwen reads MIN_PIXELS /",
+    "MAX_PIXELS from the environment and clamps each image's AREA into that band,",
+    "so it decides how many visual tokens the model reads a puzzle from. One",
+    "visual token is a 32x32 px square (patch_size 16, then spatial_merge_size 2).",
+    "It lives here, per model, for exactly the reason prompt_version does — one",
+    "client constant cannot be right for two adapters with different budgets.",
+    "",
+    "SERVE-AT, not trained-at. Seed it from the training run, then tune it by",
+    "measurement: these adapters generalize UPWARD, so the best serving band is",
+    "usually larger than the training floor. Serving under a band nobody checked",
+    "is the prompt bug in a different costume — nothing errors, the model just",
+    "answers on a picture at a scale it was not tuned for.",
+    "",
+    "Omit it and the client keeps its historical behaviour (floor at 448 squared,",
+    "no ceiling). Absent means UNKNOWN, not zero — set it only where it has been",
+    "measured or the training run is known."
+  ],
+  "latest": "CaptchaKraken/CaptchaKraken-Lora-v1.2",
+  "models": {
+    "CaptchaKraken/CaptchaKraken-Lora-v1.2": {
+      "prompt_version": "2",
+      "kind": "lora",
+      "base_model": "RedHatAI/Qwen3.5-9B-FP8-dynamic",
+      "lora_revision": "main",
+      "lora_name": "captcha-v12",
+      "pixel_budget": {
+        "min": 518400,
+        "max": 518400
+      },
+      "note": "Run 20260812-005302 (ready-to-deploy). PUBLIC on the Hub under Source-Available License v1.1, and the first published generation-2 model. TRAINED with floor 448^2 / ceiling 720^2, but MEASURED best served at a flat 720^2 — floor == ceiling, so every image arrives at exactly one size. Tier 2 overall 0.6115 -> 0.6412 across the full 1458-sample real eval and reCAPTCHA 3x3 exact 0.648 -> 0.715, against run-to-run noise of +/-0.001. Swept 2026-08-18 over 448/576/640/704/720/736/768/896/1024^2: the plateau is 704-736^2 and it falls off above 768^2, where a 400x580 portrait capture runs more than 40% past the ViT's 48x48 position grid. Note 640^2 is where that grid is exactly saturated and it scores WORSE (0.6357) — mild extrapolation plus real pixels beats staying inside the grid. Was served as `captcha-v12` while absent from this file, which resolved it to generation 1 prompts against generation-2 weights — the failure this file exists to prevent."
+    },
+    "CaptchaKraken/Twilight-v1.2-FP8": {
+      "prompt_version": "2",
+      "kind": "merged",
+      "lora_name": "CaptchaKraken/Twilight-v1.2-FP8",
+      "pixel_budget": {
+        "min": 518400,
+        "max": 518400
+      },
+      "note": "CaptchaKraken-Lora-v1.2 merged into Qwen/Qwen3.5-9B at FP8-dynamic. The weights the hosted API answers with. Generation 2, same band as the adapter it is made from: a merge that resolved to a different prompt generation or a different pixel budget than its own LoRA would be the 2026-07-18 regression with extra steps."
+    },
+    "CaptchaKraken/Sunlight-v1.2-AWQ-4bit": {
+      "prompt_version": "2",
+      "kind": "merged",
+      "lora_name": "CaptchaKraken/Sunlight-v1.2-AWQ-4bit",
+      "pixel_budget": {
+        "min": 518400,
+        "max": 518400
+      },
+      "note": "CaptchaKraken-Lora-v1.2 merged into Qwen/Qwen3.5-9B at AWQ 4-bit. Generation 2. Quantization changes the weights, not the schema they answer in, so the prompt generation and the pixel budget are the adapter's."
+    },
+    "CaptchaKraken/CaptchaKraken_v1.1": {
+      "prompt_version": "1",
+      "kind": "lora",
+      "base_model": "RedHatAI/Qwen3.5-9B-FP8-dynamic",
+      "lora_revision": "main",
+      "lora_name": "captcha",
+      "note": "Currently served in production. PRIVATE on the Hub, so prompts.json can never be fetched for it — this entry is the only thing that maps it to a prompt generation."
+    },
+    "CaptchaKraken/CaptchaKrakenV1_Lora": {
+      "prompt_version": "1",
+      "kind": "lora",
+      "base_model": "RedHatAI/Qwen3.5-9B-FP8-dynamic",
+      "lora_revision": "main",
+      "lora_name": "captcha",
+      "note": "Published before prompts.json existed — no such file in the repo, so it resolves from here."
+    },
+    "CaptchaKraken/Sunlight-AWQ-4bit": {
+      "prompt_version": "1",
+      "kind": "merged",
+      "lora_name": "captcha",
+      "note": "Ships its own prompts.json; this entry keeps it resolvable offline."
+    },
+    "CaptchaKraken/Twilight-FP8": {
+      "prompt_version": "1",
+      "kind": "merged",
+      "lora_name": "captcha",
+      "note": "Ships its own prompts.json; this entry keeps it resolvable offline."
+    }
+  },
+  "served_aliases": {
+    "_comment": "A hosted endpoint serves a bare name, not a repo id. Map it back so prompt resolution works for callers who only ever see the alias.",
+    "captcha": "CaptchaKraken/CaptchaKraken_v1.1",
+    "captcha-v12": "CaptchaKraken/CaptchaKraken-Lora-v1.2"
+  }
+}