offgrid-ai 0.12.1 → 0.12.2

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.
Files changed (2) hide show
  1. package/package.json +1 -1
  2. package/src/mlx-flags.mjs +7 -0
package/package.json CHANGED
@@ -1,6 +1,6 @@
1
1
  {
2
2
  "name": "offgrid-ai",
3
- "version": "0.12.1",
3
+ "version": "0.12.2",
4
4
  "description": "Privacy-first CLI for running local LLMs — discover, configure, run, benchmark",
5
5
  "author": "Eeshan Srivastava (https://eeshans.com)",
6
6
  "type": "module",
package/src/mlx-flags.mjs CHANGED
@@ -89,5 +89,12 @@ export function computeMlxVlmFlags(modelPath, options = {}) {
89
89
  args.push("--max-kv-size", String(ctxSize));
90
90
  }
91
91
 
92
+ // Default max output tokens — used when the client doesn't specify max_tokens
93
+ // in the request. Pi's OpenAI completions provider never sends max_tokens
94
+ // (it doesn't fall back to model.maxTokens like the Anthropic provider does).
95
+ // llama-server defaults high; mlx-vlm defaults to 2048 which is too low for
96
+ // coding tasks. Set a generous server-side default.
97
+ args.push("--max-tokens", "16384");
98
+
92
99
  return { args, port };
93
100
  }