pi-blablador 0.1.2 → 0.1.3

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
package/README.md CHANGED
@@ -46,6 +46,7 @@ so the map is per model family:
46
46
  | kimi | none \| low \| high \| max |
47
47
  | deepseek | none \| low \| high \| xhigh \| max |
48
48
  | muse | low \| medium \| high \| xhigh |
49
+ | mistral (`alias-mistral-large`) | low \| medium \| high \| xhigh \| max |
49
50
  | generic (glm, minimax, mimo, aleph, …) | none \| low \| medium \| high \| xhigh |
50
51
  | none (apertus, eve, ocr) | — (the field is never sent) |
51
52
 
@@ -55,14 +56,16 @@ model sends `xhigh`, `--thinking medium` on DeepSeek sends `high`, `--thinking m
55
56
  ## Context windows
56
57
 
57
58
  The context windows the gateway enforces: 1,048,576 for the 1M deployments, 262,144 for the Qwen 3.8,
58
- Apertus and OCR families, and 131,072 for `alias-fast` and `alias-eve`. Output is capped at 16k
59
+ Apertus, OCR and Mistral families, and 131,072 for `alias-fast` and `alias-eve`. Output is capped at 16k
59
60
  tokens so reasoning cannot consume the whole answer budget.
60
61
 
61
62
  ## Which models appear
62
63
 
63
64
  `alias-*` ids, plus two verified unaliased ids (`deepseek-v4.1-flash-at-hzdr`, `Swift-Qwen3.8-27b`).
64
- The alias is the stable handle — JSC's guide says to use it so the model behind it can change. The rest
65
- of the catalog is either a descriptive deployment name ("90 - Kimi-K3 1M on Juwels Booster"), an
65
+ The alias is the stable handle — JSC's guide says to use it so the model behind it can change. Rows
66
+ read `alias (deployment)`, alias first, so what you type leads the row:
67
+ `alias-large (nvidia/Qwen3.8-Flash-Next-NVFP4)`. The rest of the catalog is either a descriptive
68
+ deployment name ("90 - Kimi-K3 1M on Juwels Booster"), an
66
69
  OpenAI-name shim for the *embedding* endpoint (`gpt-3.5-turbo`, `text-davinci-003`), or a model that
67
70
  cannot call tools (`trustllm-*`). Embeddings, rerankers, image/video/voice and the anonymizer are not
68
71
  chat models and never appear.
package/blablador.ts CHANGED
@@ -58,6 +58,9 @@ function chatModel(model: CatalogModel): ProviderModelConfig {
58
58
  // Free research API; a real rate card would make /usage meaningful.
59
59
  cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 },
60
60
  contextWindow: model.contextWindow,
61
+ // The gateway answers HTTP 400 ("Unknown message role: developer") to anything but
62
+ // system/user/assistant/tool, so pi must never use the newer role here.
63
+ compat: { supportsDeveloperRole: false },
61
64
  // Reasoning shares the output budget; outputCap keeps the prompt room inside the window
62
65
  // even if the gateway ever reports one smaller than the ceiling (README "Context windows").
63
66
  maxTokens: outputCap(model.contextWindow),
package/package.json CHANGED
@@ -1,6 +1,6 @@
1
1
  {
2
2
  "name": "pi-blablador",
3
- "version": "0.1.2",
3
+ "version": "0.1.3",
4
4
  "description": "Blablador (Helmholtz / JSC) provider for Pi: live alias discovery, per-family thinking effort, measured context windows",
5
5
  "type": "module",
6
6
  "license": "MIT",
@@ -34,7 +34,7 @@
34
34
  "node": ">=22.18.0"
35
35
  },
36
36
  "dependencies": {
37
- "blablador-core": "0.1.2"
37
+ "blablador-core": "0.1.3"
38
38
  },
39
39
  "scripts": {
40
40
  "test": "node --test \"test/**/*.test.mjs\"",