@selesai/code 0.5.24 → 0.5.26

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 (55) hide show
  1. package/dist/extensions/pi-powerline-footer/fixed-editor/terminal-split.ts +4 -1
  2. package/dist/extensions/pi-powerline-footer/index.ts +1 -0
  3. package/dist/extensions/workflow/adapter.ts +2 -2
  4. package/dist/extensions/workflow/modes/loop.ts +30 -11
  5. package/docs/workflows.md +3 -3
  6. package/package.json +2 -2
  7. package/dist/core/provider-display-names.d.ts +0 -2
  8. package/dist/core/provider-display-names.d.ts.map +0 -1
  9. package/dist/core/provider-display-names.js +0 -36
  10. package/dist/core/provider-display-names.js.map +0 -1
  11. package/dist/data/capabilities.d.ts +0 -48
  12. package/dist/data/capabilities.d.ts.map +0 -1
  13. package/dist/data/capabilities.js +0 -282
  14. package/dist/data/capabilities.js.map +0 -1
  15. package/dist/data/navigation.d.ts +0 -3
  16. package/dist/data/navigation.d.ts.map +0 -1
  17. package/dist/data/navigation.js +0 -90
  18. package/dist/data/navigation.js.map +0 -1
  19. package/dist/extensions/index.d.ts +0 -3
  20. package/dist/extensions/index.d.ts.map +0 -1
  21. package/dist/extensions/index.js +0 -9
  22. package/dist/extensions/index.js.map +0 -1
  23. package/dist/extensions/llama/client.d.ts +0 -61
  24. package/dist/extensions/llama/client.d.ts.map +0 -1
  25. package/dist/extensions/llama/client.js +0 -309
  26. package/dist/extensions/llama/client.js.map +0 -1
  27. package/dist/extensions/llama/huggingface.d.ts +0 -23
  28. package/dist/extensions/llama/huggingface.d.ts.map +0 -1
  29. package/dist/extensions/llama/huggingface.js +0 -146
  30. package/dist/extensions/llama/huggingface.js.map +0 -1
  31. package/dist/extensions/llama/index.d.ts +0 -3
  32. package/dist/extensions/llama/index.d.ts.map +0 -1
  33. package/dist/extensions/llama/index.js +0 -211
  34. package/dist/extensions/llama/index.js.map +0 -1
  35. package/dist/extensions/llama/provider.d.ts +0 -10
  36. package/dist/extensions/llama/provider.d.ts.map +0 -1
  37. package/dist/extensions/llama/provider.js +0 -106
  38. package/dist/extensions/llama/provider.js.map +0 -1
  39. package/dist/extensions/llama/ui.d.ts +0 -42
  40. package/dist/extensions/llama/ui.d.ts.map +0 -1
  41. package/dist/extensions/llama/ui.js +0 -420
  42. package/dist/extensions/llama/ui.js.map +0 -1
  43. package/dist/extensions/pi-intercom/lifecycle.test.ts +0 -60
  44. package/dist/extensions/pi-subagents/src/runs/background/wait.ts +0 -353
  45. package/dist/extensions/pi-subagents/test/unit/live-async-resume.test.ts +0 -170
  46. package/dist/extensions/pi-subagents/test/unit/wait.test.ts +0 -398
  47. package/dist/extensions/question/dialog-adapter.ts +0 -96
  48. package/dist/extensions/question/tests/ui-protocol.test.ts +0 -157
  49. package/dist/extensions/question/tui-adapter.ts +0 -15
  50. package/dist/extensions/question/ui-protocol.ts +0 -31
  51. package/dist/extensions/workflow/modes/quick.ts +0 -150
  52. package/dist/modes/interactive/interactive-mode.test.d.ts +0 -2
  53. package/dist/modes/interactive/interactive-mode.test.d.ts.map +0 -1
  54. package/dist/modes/interactive/interactive-mode.test.js +0 -45
  55. package/dist/modes/interactive/interactive-mode.test.js.map +0 -1
@@ -1 +0,0 @@
1
- {"version":3,"file":"client.js","sourceRoot":"","sources":["../../../src/extensions/llama/client.ts"],"names":[],"mappings":";;;;;;AA0CA,SAAS,YAAY,CAAC,OAAgB,EAAE,QAAgB,EAAU;IACjE,IAAI,OAAO,OAAO,KAAK,QAAQ,IAAI,OAAO,KAAK,IAAI;QAAE,OAAO,QAAQ,CAAC;IACrE,MAAM,KAAK,GAAI,OAA+B,CAAC,KAAK,CAAC;IACrD,IAAI,OAAO,KAAK,KAAK,QAAQ,IAAI,KAAK,KAAK,IAAI;QAAE,OAAO,QAAQ,CAAC;IACjE,MAAM,OAAO,GAAI,KAA+B,CAAC,OAAO,CAAC;IACzD,OAAO,OAAO,OAAO,KAAK,QAAQ,IAAI,OAAO,CAAC,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,QAAQ,CAAC;AAAA,CACnE;AAED,SAAS,WAAW,CAAC,KAAc,EAA2B;IAC7D,IAAI,OAAO,KAAK,KAAK,QAAQ,IAAI,KAAK,KAAK,IAAI;QAAE,OAAO,KAAK,CAAC;IAC9D,MAAM,SAAS,GAAG,KAAuD,CAAC;IAC1E,OAAO,OAAO,SAAS,CAAC,EAAE,KAAK,QAAQ,IAAI,OAAO,SAAS,CAAC,MAAM,EAAE,KAAK,KAAK,QAAQ,CAAC;AAAA,CACvF;AAED,SAAS,UAAU,CAAC,MAA+B,EAAE,MAAuB,EAAc;IACzF,IAAI,CAAC,MAAM;QAAE,OAAO,GAAG,EAAE,CAAC,EAAC,CAAC,CAAC;IAC7B,IAAI,MAAM,CAAC,OAAO,EAAE,CAAC;QACpB,MAAM,CAAC,KAAK,CAAC,MAAM,CAAC,MAAM,CAAC,CAAC;QAC5B,OAAO,GAAG,EAAE,CAAC,EAAC,CAAC,CAAC;IACjB,CAAC;IACD,MAAM,KAAK,GAAG,GAAG,EAAE,CAAC,MAAM,CAAC,KAAK,CAAC,MAAM,CAAC,MAAM,CAAC,CAAC;IAChD,MAAM,CAAC,gBAAgB,CAAC,OAAO,EAAE,KAAK,EAAE,EAAE,IAAI,EAAE,IAAI,EAAE,CAAC,CAAC;IACxD,OAAO,GAAG,EAAE,CAAC,MAAM,CAAC,mBAAmB,CAAC,OAAO,EAAE,KAAK,CAAC,CAAC;AAAA,CACxD;AAED,SAAS,KAAK,CAAC,EAAU,EAAE,MAAoB,EAAiB;IAC/D,OAAO,IAAI,OAAO,CAAC,CAAC,OAAO,EAAE,MAAM,EAAE,EAAE,CAAC;QACvC,IAAI,MAAM,EAAE,OAAO,EAAE,CAAC;YACrB,MAAM,CAAC,MAAM,CAAC,MAAM,IAAI,IAAI,KAAK,CAAC,WAAW,CAAC,CAAC,CAAC;YAChD,OAAO;QACR,CAAC;QACD,MAAM,KAAK,GAAG,GAAG,EAAE,CAAC;YACnB,YAAY,CAAC,OAAO,CAAC,CAAC;YACtB,MAAM,CAAC,MAAM,EAAE,MAAM,IAAI,IAAI,KAAK,CAAC,WAAW,CAAC,CAAC,CAAC;QAAA,CACjD,CAAC;QACF,MAAM,OAAO,GAAG,UAAU,CAAC,GAAG,EAAE,CAAC;YAChC,MAAM,EAAE,mBAAmB,CAAC,OAAO,EAAE,KAAK,CAAC,CAAC;YAC5C,OAAO,EAAE,CAAC;QAAA,CACV,EAAE,EAAE,CAAC,CAAC;QACP,MAAM,EAAE,gBAAgB,CAAC,OAAO,EAAE,KAAK,EAAE,EAAE,IAAI,EAAE,IAAI,EAAE,CAAC,CAAC;IAAA,CACzD,CAAC,CAAC;AAAA,CACH;AAED,SAAS,iBAAiB,CAAC,IAAa,EAA6B;IACpE,IAAI,OAAO,IAAI,KAAK,QAAQ,IAAI,IAAI,KAAK,IAAI;QAAE,OAAO,SAAS,CAAC;IAChE,MAAM,QAAQ,GAAI,IAA+B,CAAC,QAAQ,CAAC;IAC3D,IAAI,OAAO,QAAQ,KAAK,QAAQ,IAAI,QAAQ,KAAK,IAAI;QAAE,OAAO,SAAS,CAAC;IACxE,MAAM,KAAK,GAAG,QAAqF,CAAC;IACpG,MAAM,KAAK,GACV,OAAO,KAAK,CAAC,OAAO,KAAK,QAAQ,CAAC,CAAC,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,OAAO,KAAK,CAAC,KAAK,KAAK,QAAQ,CAAC,CAAC,CAAC,KAAK,CAAC,KAAK,CAAC,CAAC,CAAC,SAAS,CAAC;IAC/G,MAAM,MAAM,GAAG,KAAK,CAAC,OAAO,CAAC,KAAK,CAAC,MAAM,CAAC;QACzC,CAAC,CAAC,KAAK,CAAC,MAAM,CAAC,MAAM,CAAC,CAAC,KAAK,EAAmB,EAAE,CAAC,OAAO,KAAK,KAAK,QAAQ,CAAC;QAC5E,CAAC,CAAC,EAAE,CAAC;IACN,MAAM,UAAU,GAAG,OAAO,KAAK,CAAC,KAAK,KAAK,QAAQ,CAAC,CAAC,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,KAAK,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC;IACvG,IAAI,KAAK,GAAG,UAAU,CAAC;IACvB,IAAI,KAAK,IAAI,MAAM,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;QAChC,MAAM,KAAK,GAAG,MAAM,CAAC,OAAO,CAAC,KAAK,CAAC,CAAC;QACpC,IAAI,KAAK,IAAI,CAAC;YAAE,KAAK,GAAG,CAAC,KAAK,GAAG,CAAC,UAAU,IAAI,CAAC,CAAC,CAAC,GAAG,MAAM,CAAC,MAAM,CAAC;IACrE,CAAC;IACD,OAAO;QACN,OAAO,EAAE,KAAK,CAAC,CAAC,CAAC,WAAW,KAAK,CAAC,UAAU,CAAC,GAAG,EAAE,GAAG,CAAC,EAAE,CAAC,CAAC,CAAC,eAAe;QAC1E,KAAK;KACL,CAAC;AAAA,CACF;AAED,SAAS,qBAAqB,CAAC,IAAa,EAA6B;IACxE,IAAI,OAAO,IAAI,KAAK,QAAQ,IAAI,IAAI,KAAK,IAAI;QAAE,OAAO,SAAS,CAAC;IAChE,MAAM,MAAM,GAAI,IAA+B,CAAC,QAAQ,CAAC;IACzD,MAAM,KAAK,GAAG,OAAO,MAAM,KAAK,QAAQ,IAAI,MAAM,KAAK,IAAI,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,IAAI,CAAC;IAC5E,IAAI,IAAI,GAAG,CAAC,CAAC;IACb,IAAI,KAAK,GAAG,CAAC,CAAC;IACd,KAAK,MAAM,KAAK,IAAI,MAAM,CAAC,MAAM,CAAC,KAAgC,CAAC,EAAE,CAAC;QACrE,IAAI,OAAO,KAAK,KAAK,QAAQ,IAAI,KAAK,KAAK,IAAI;YAAE,SAAS;QAC1D,MAAM,KAAK,GAAG,KAA4C,CAAC;QAC3D,IAAI,OAAO,KAAK,CAAC,IAAI,KAAK,QAAQ,IAAI,OAAO,KAAK,CAAC,KAAK,KAAK,QAAQ;YAAE,SAAS;QAChF,IAAI,IAAI,KAAK,CAAC,IAAI,CAAC;QACnB,KAAK,IAAI,KAAK,CAAC,KAAK,CAAC;IACtB,CAAC;IACD,IAAI,KAAK,IAAI,CAAC;QAAE,OAAO,SAAS,CAAC;IACjC,OAAO;QACN,OAAO,EAAE,mBAAmB;QAC5B,KAAK,EAAE,IAAI,GAAG,KAAK;QACnB,MAAM,EAAE,GAAG,WAAW,CAAC,IAAI,CAAC,MAAM,WAAW,CAAC,KAAK,CAAC,EAAE;KACtD,CAAC;AAAA,CACF;AAED,qBAA4B,KAAa,EAAU;IAClD,IAAI,KAAK,GAAG,IAAI;QAAE,OAAO,GAAG,KAAK,IAAI,CAAC;IACtC,MAAM,KAAK,GAAG,CAAC,KAAK,EAAE,KAAK,EAAE,KAAK,EAAE,KAAK,CAAC,CAAC;IAC3C,IAAI,KAAK,GAAG,KAAK,GAAG,IAAI,CAAC;IACzB,IAAI,IAAI,GAAG,KAAK,CAAC,CAAC,CAAE,CAAC;IACrB,KAAK,IAAI,KAAK,GAAG,CAAC,EAAE,KAAK,GAAG,KAAK,CAAC,MAAM,IAAI,KAAK,IAAI,IAAI,EAAE,KAAK,EAAE,EAAE,CAAC;QACpE,KAAK,IAAI,IAAI,CAAC;QACd,IAAI,GAAG,KAAK,CAAC,KAAK,CAAE,CAAC;IACtB,CAAC;IACD,OAAO,GAAG,KAAK,IAAI,EAAE,CAAC,CAAC,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,IAAI,IAAI,EAAE,CAAC;AAAA,CACtE;AAED,iCAAwC,KAAa,EAAU;IAC9D,MAAM,GAAG,GAAG,IAAI,GAAG,CAAC,KAAK,CAAC,IAAI,EAAE,CAAC,CAAC;IAClC,IAAI,GAAG,CAAC,QAAQ,KAAK,OAAO,IAAI,GAAG,CAAC,QAAQ,KAAK,QAAQ,EAAE,CAAC;QAC3D,MAAM,IAAI,KAAK,CAAC,mCAAmC,CAAC,CAAC;IACtD,CAAC;IACD,GAAG,CAAC,IAAI,GAAG,EAAE,CAAC;IACd,GAAG,CAAC,MAAM,GAAG,EAAE,CAAC;IAChB,GAAG,CAAC,QAAQ,GAAG,GAAG,CAAC,QAAQ,CAAC,OAAO,CAAC,OAAO,EAAE,EAAE,CAAC,CAAC,OAAO,CAAC,QAAQ,EAAE,EAAE,CAAC,IAAI,GAAG,CAAC;IAC9E,OAAO,GAAG,CAAC,QAAQ,EAAE,CAAC,OAAO,CAAC,MAAM,EAAE,EAAE,CAAC,CAAC;AAAA,CAC1C;AAED,2BAAkC,SAAiB,EAAU;IAC5D,OAAO,GAAG,uBAAuB,CAAC,SAAS,CAAC,KAAK,CAAC;AAAA,CAClD;AAED;IACU,SAAS,CAAS;IACV,MAAM,CAAqB;IAE5C,YAAY,SAAiB,EAAE,MAAe,EAAE;QAC/C,IAAI,CAAC,SAAS,GAAG,uBAAuB,CAAC,SAAS,CAAC,CAAC;QACpD,IAAI,CAAC,MAAM,GAAG,MAAM,CAAC;IAAA,CACrB;IAEO,KAAK,CAAC,OAAO,CAAC,IAAY,EAAE,IAAI,GAAgB,EAAE,EAAoB;QAC7E,MAAM,OAAO,GAAG,IAAI,OAAO,CAAC,IAAI,CAAC,OAAO,CAAC,CAAC;QAC1C,IAAI,IAAI,CAAC,IAAI,KAAK,SAAS;YAAE,OAAO,CAAC,GAAG,CAAC,cAAc,EAAE,kBAAkB,CAAC,CAAC;QAC7E,IAAI,IAAI,CAAC,MAAM;YAAE,OAAO,CAAC,GAAG,CAAC,eAAe,EAAE,UAAU,IAAI,CAAC,MAAM,EAAE,CAAC,CAAC;QACvE,MAAM,OAAO,GAAG,WAAW,CAAC,OAAO,CAAC,MAAM,CAAC,CAAC;QAC5C,MAAM,MAAM,GAAG,IAAI,CAAC,MAAM,CAAC,CAAC,CAAC,WAAW,CAAC,GAAG,CAAC,CAAC,IAAI,CAAC,MAAM,EAAE,OAAO,CAAC,CAAC,CAAC,CAAC,CAAC,OAAO,CAAC;QAC/E,MAAM,QAAQ,GAAG,MAAM,KAAK,CAAC,GAAG,IAAI,CAAC,SAAS,GAAG,IAAI,EAAE,EAAE,EAAE,GAAG,IAAI,EAAE,OAAO,EAAE,MAAM,EAAE,CAAC,CAAC;QACvF,IAAI,OAAgB,CAAC;QACrB,IAAI,CAAC;YACJ,OAAO,GAAG,MAAM,QAAQ,CAAC,IAAI,EAAE,CAAC;QACjC,CAAC;QAAC,MAAM,CAAC;YACR,OAAO,GAAG,SAAS,CAAC;QACrB,CAAC;QACD,IAAI,CAAC,QAAQ,CAAC,EAAE;YAAE,MAAM,IAAI,KAAK,CAAC,YAAY,CAAC,OAAO,EAAE,2BAA2B,QAAQ,CAAC,MAAM,EAAE,CAAC,CAAC,CAAC;QACvG,OAAO,OAAO,CAAC;IAAA,CACf;IAED,KAAK,CAAC,IAAI,CAAC,OAAO,GAA+C,EAAE,EAA6B;QAC/F,MAAM,OAAO,GAAG,MAAM,IAAI,CAAC,OAAO,CAAC,UAAU,OAAO,CAAC,MAAM,CAAC,CAAC,CAAC,WAAW,CAAC,CAAC,CAAC,EAAE,EAAE,EAAE,EAAE,MAAM,EAAE,OAAO,CAAC,MAAM,EAAE,CAAC,CAAC;QAC9G,IAAI,OAAO,OAAO,KAAK,QAAQ,IAAI,OAAO,KAAK,IAAI,IAAI,CAAC,KAAK,CAAC,OAAO,CAAE,OAA8B,CAAC,IAAI,CAAC,EAAE,CAAC;YAC7G,MAAM,IAAI,KAAK,CAAC,6CAA6C,CAAC,CAAC;QAChE,CAAC;QACD,MAAM,IAAI,GAAI,OAA+B,CAAC,IAAI,CAAC;QACnD,IAAI,CAAC,IAAI,CAAC,KAAK,CAAC,WAAW,CAAC;YAAE,MAAM,IAAI,KAAK,CAAC,gDAAgD,CAAC,CAAC;QAChG,OAAO,IAAI,CAAC;IAAA,CACZ;IAED,KAAK,CAAC,IAAI,CAAC,KAAa,EAAE,MAAoB,EAAiB;QAC9D,MAAM,IAAI,CAAC,OAAO,CAAC,cAAc,EAAE,EAAE,MAAM,EAAE,MAAM,EAAE,IAAI,EAAE,IAAI,CAAC,SAAS,CAAC,EAAE,KAAK,EAAE,CAAC,EAAE,MAAM,EAAE,CAAC,CAAC;IAAA,CAChG;IAED,KAAK,CAAC,MAAM,CAAC,KAAa,EAAE,MAAoB,EAAiB;QAChE,MAAM,IAAI,CAAC,OAAO,CAAC,gBAAgB,EAAE,EAAE,MAAM,EAAE,MAAM,EAAE,IAAI,EAAE,IAAI,CAAC,SAAS,CAAC,EAAE,KAAK,EAAE,CAAC,EAAE,MAAM,EAAE,CAAC,CAAC;IAAA,CAClG;IAED,KAAK,CAAC,aAAa,CAAC,KAAa,EAAE,MAAoB,EAAiB;QACvE,MAAM,IAAI,CAAC,MAAM,CAAC,KAAK,EAAE,MAAM,CAAC,CAAC;QACjC,OAAO,IAAI,EAAE,CAAC;YACb,MAAM,KAAK,GAAG,CAAC,MAAM,IAAI,CAAC,IAAI,CAAC,EAAE,MAAM,EAAE,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,SAAS,EAAE,EAAE,CAAC,SAAS,CAAC,EAAE,KAAK,KAAK,CAAC,CAAC;YACxF,IAAI,CAAC,KAAK,IAAI,KAAK,CAAC,MAAM,CAAC,KAAK,KAAK,UAAU;gBAAE,OAAO;YACxD,MAAM,KAAK,CAAC,GAAG,EAAE,MAAM,CAAC,CAAC;QAC1B,CAAC;IAAA,CACD;IAED,KAAK,CAAC,QAAQ,CAAC,KAAa,EAAE,MAAoB,EAAiB;QAClE,MAAM,IAAI,CAAC,OAAO,CAAC,SAAS,EAAE,EAAE,MAAM,EAAE,MAAM,EAAE,IAAI,EAAE,IAAI,CAAC,SAAS,CAAC,EAAE,KAAK,EAAE,CAAC,EAAE,MAAM,EAAE,CAAC,CAAC;IAAA,CAC3F;IAED,KAAK,CAAC,KAAK,CAAC,OAAyC,EAAE,MAAoB,EAAiB;QAC3F,MAAM,OAAO,GAAG,IAAI,OAAO,EAAE,CAAC;QAC9B,IAAI,IAAI,CAAC,MAAM;YAAE,OAAO,CAAC,GAAG,CAAC,eAAe,EAAE,UAAU,IAAI,CAAC,MAAM,EAAE,CAAC,CAAC;QACvE,MAAM,QAAQ,GAAG,MAAM,KAAK,CAAC,GAAG,IAAI,CAAC,SAAS,aAAa,EAAE,EAAE,OAAO,EAAE,MAAM,EAAE,CAAC,CAAC;QAClF,IAAI,CAAC,QAAQ,CAAC,EAAE,IAAI,CAAC,QAAQ,CAAC,IAAI;YAAE,MAAM,IAAI,KAAK,CAAC,+BAA+B,QAAQ,CAAC,MAAM,EAAE,CAAC,CAAC;QACtG,MAAM,MAAM,GAAG,QAAQ,CAAC,IAAI,CAAC,SAAS,EAAE,CAAC;QACzC,MAAM,OAAO,GAAG,IAAI,WAAW,EAAE,CAAC;QAClC,IAAI,MAAM,GAAG,EAAE,CAAC;QAChB,OAAO,IAAI,EAAE,CAAC;YACb,MAAM,KAAK,GAAG,MAAM,MAAM,CAAC,IAAI,EAAE,CAAC;YAClC,IAAI,KAAK,CAAC,IAAI;gBAAE,MAAM;YACtB,MAAM,IAAI,OAAO,CAAC,MAAM,CAAC,KAAK,CAAC,KAAK,EAAE,EAAE,MAAM,EAAE,IAAI,EAAE,CAAC,CAAC,UAAU,CAAC,MAAM,EAAE,IAAI,CAAC,CAAC;YACjF,IAAI,QAAQ,GAAG,MAAM,CAAC,OAAO,CAAC,MAAM,CAAC,CAAC;YACtC,OAAO,QAAQ,IAAI,CAAC,EAAE,CAAC;gBACtB,MAAM,KAAK,GAAG,MAAM,CAAC,KAAK,CAAC,CAAC,EAAE,QAAQ,CAAC,CAAC;gBACxC,MAAM,GAAG,MAAM,CAAC,KAAK,CAAC,QAAQ,GAAG,CAAC,CAAC,CAAC;gBACpC,MAAM,IAAI,GAAG,KAAK;qBAChB,KAAK,CAAC,IAAI,CAAC;qBACX,MAAM,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,CAAC,UAAU,CAAC,OAAO,CAAC,CAAC;qBAC1C,GAAG,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,SAAS,EAAE,CAAC;qBACxC,IAAI,CAAC,IAAI,CAAC,CAAC;gBACb,IAAI,IAAI,EAAE,CAAC;oBACV,IAAI,CAAC;wBACJ,MAAM,KAAK,GAAG,IAAI,CAAC,KAAK,CAAC,IAAI,CAAoB,CAAC;wBAClD,IAAI,KAAK,IAAI,OAAO,KAAK,CAAC,KAAK,KAAK,QAAQ,IAAI,OAAO,KAAK,CAAC,KAAK,KAAK,QAAQ;4BAAE,OAAO,CAAC,KAAK,CAAC,CAAC;oBACjG,CAAC;oBAAC,MAAM,CAAC;wBACR,kEAAkE;oBACnE,CAAC;gBACF,CAAC;gBACD,QAAQ,GAAG,MAAM,CAAC,OAAO,CAAC,MAAM,CAAC,CAAC;YACnC,CAAC;QACF,CAAC;IAAA,CACD;IAED,KAAK,CAAC,WAAW,CAChB,KAAa,EACb,UAA6C,EAC7C,MAAoB,EACM;QAC1B,MAAM,OAAO,GAAG,IAAI,eAAe,EAAE,CAAC;QACtC,MAAM,MAAM,GAAG,UAAU,CAAC,MAAM,EAAE,OAAO,CAAC,CAAC;QAC3C,IAAI,WAAW,GAAG,KAAK,CAAC;QACxB,IAAI,UAA8B,CAAC;QACnC,KAAK,IAAI,CAAC,KAAK,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC;YAC1B,IAAI,KAAK,CAAC,KAAK,KAAK,KAAK;gBAAE,OAAO;YAClC,IAAI,KAAK,CAAC,KAAK,KAAK,cAAc,IAAI,KAAK,CAAC,KAAK,KAAK,eAAe;gBAAE,OAAO;YAC9E,MAAM,IAAI,GAAG,KAAK,CAAC,IAAwC,CAAC;YAC5D,IAAI,IAAI,EAAE,MAAM,KAAK,QAAQ;gBAAE,WAAW,GAAG,IAAI,CAAC;YAClD,IAAI,IAAI,EAAE,MAAM,KAAK,UAAU;gBAAE,UAAU,GAAG,sBAAsB,CAAC;YACrE,MAAM,QAAQ,GAAG,iBAAiB,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC;YAC/C,IAAI,QAAQ;gBAAE,UAAU,CAAC,QAAQ,CAAC,CAAC;QAAA,CACnC,EAAE,OAAO,CAAC,MAAM,CAAC,CAAC,KAAK,CAAC,GAAG,EAAE,CAAC,EAAC,CAAC,CAAC,CAAC;QACnC,IAAI,CAAC;YACJ,MAAM,IAAI,CAAC,IAAI,CAAC,KAAK,EAAE,MAAM,CAAC,CAAC;YAC/B,UAAU,CAAC,EAAE,OAAO,EAAE,eAAe,EAAE,CAAC,CAAC;YACzC,OAAO,IAAI,EAAE,CAAC;gBACb,IAAI,MAAM,EAAE,OAAO;oBAAE,MAAM,MAAM,CAAC,MAAM,IAAI,IAAI,KAAK,CAAC,WAAW,CAAC,CAAC;gBACnE,MAAM,KAAK,GAAG,CAAC,MAAM,IAAI,CAAC,IAAI,CAAC,EAAE,MAAM,EAAE,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,SAAS,EAAE,EAAE,CAAC,SAAS,CAAC,EAAE,KAAK,KAAK,CAAC,CAAC;gBACxF,IAAI,KAAK,EAAE,MAAM,CAAC,KAAK,KAAK,QAAQ;oBAAE,OAAO,KAAK,CAAC;gBACnD,IAAI,WAAW,IAAI,CAAC,KAAK;oBAAE,OAAO,EAAE,EAAE,EAAE,KAAK,EAAE,MAAM,EAAE,EAAE,KAAK,EAAE,QAAQ,EAAE,EAAE,CAAC;gBAC7E,IAAI,KAAK,EAAE,MAAM,CAAC,MAAM,IAAI,UAAU,EAAE,CAAC;oBACxC,MAAM,IAAI,KAAK,CACd,KAAK,EAAE,MAAM,CAAC,SAAS,KAAK,SAAS;wBACpC,CAAC,CAAC,CAAC,UAAU,IAAI,sBAAsB,CAAC;wBACxC,CAAC,CAAC,0BAA0B,KAAK,CAAC,MAAM,CAAC,SAAS,EAAE,CACrD,CAAC;gBACH,CAAC;gBACD,MAAM,KAAK,CAAC,GAAG,EAAE,MAAM,CAAC,CAAC;YAC1B,CAAC;QACF,CAAC;gBAAS,CAAC;YACV,MAAM,EAAE,CAAC;YACT,OAAO,CAAC,KAAK,EAAE,CAAC;QACjB,CAAC;IAAA,CACD;IAED,KAAK,CAAC,eAAe,CACpB,KAAa,EACb,UAA6C,EAC7C,MAAoB,EACQ;QAC5B,MAAM,OAAO,GAAG,IAAI,eAAe,EAAE,CAAC;QACtC,MAAM,MAAM,GAAG,UAAU,CAAC,MAAM,EAAE,OAAO,CAAC,CAAC;QAC3C,IAAI,QAAQ,GAAG,KAAK,CAAC;QACrB,IAAI,OAA2B,CAAC;QAChC,IAAI,cAAc,GAAG,KAAK,CAAC;QAC3B,IAAI,KAAK,GAAG,CAAC,CAAC;QACd,KAAK,IAAI,CAAC,KAAK,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC;YAC1B,IAAI,KAAK,CAAC,KAAK,KAAK,KAAK;gBAAE,OAAO;YAClC,IAAI,KAAK,CAAC,KAAK,KAAK,mBAAmB;gBAAE,QAAQ,GAAG,IAAI,CAAC;YACzD,IAAI,KAAK,CAAC,KAAK,KAAK,iBAAiB;gBAAE,OAAO,GAAG,YAAY,CAAC,KAAK,CAAC,IAAI,EAAE,iBAAiB,CAAC,CAAC;YAC7F,IAAI,KAAK,CAAC,KAAK,KAAK,mBAAmB,EAAE,CAAC;gBACzC,cAAc,GAAG,IAAI,CAAC;gBACtB,MAAM,QAAQ,GAAG,qBAAqB,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC;gBACnD,IAAI,QAAQ;oBAAE,UAAU,CAAC,QAAQ,CAAC,CAAC;YACpC,CAAC;QAAA,CACD,EAAE,OAAO,CAAC,MAAM,CAAC,CAAC,KAAK,CAAC,GAAG,EAAE,CAAC,EAAC,CAAC,CAAC,CAAC;QACnC,IAAI,CAAC;YACJ,MAAM,IAAI,CAAC,QAAQ,CAAC,KAAK,EAAE,MAAM,CAAC,CAAC;YACnC,UAAU,CAAC,EAAE,OAAO,EAAE,mBAAmB,EAAE,CAAC,CAAC;YAC7C,OAAO,IAAI,EAAE,CAAC;gBACb,IAAI,MAAM,EAAE,OAAO;oBAAE,MAAM,MAAM,CAAC,MAAM,IAAI,IAAI,KAAK,CAAC,WAAW,CAAC,CAAC;gBACnE,IAAI,OAAO;oBAAE,MAAM,IAAI,KAAK,CAAC,OAAO,CAAC,CAAC;gBACtC,MAAM,MAAM,GAAG,MAAM,IAAI,CAAC,IAAI,CAAC,EAAE,MAAM,EAAE,CAAC,CAAC;gBAC3C,KAAK,EAAE,CAAC;gBACR,MAAM,KAAK,GAAG,MAAM,CAAC,IAAI,CAAC,CAAC,SAAS,EAAE,EAAE,CAAC,SAAS,CAAC,EAAE,KAAK,KAAK,CAAC,CAAC;gBACjE,IAAI,KAAK,EAAE,MAAM,CAAC,KAAK,KAAK,aAAa,EAAE,CAAC;oBAC3C,cAAc,GAAG,IAAI,CAAC;oBACtB,MAAM,QAAQ,GAAG,qBAAqB,CAAC,KAAK,CAAC,MAAM,CAAC,QAAQ,CAAC,CAAC;oBAC9D,IAAI,QAAQ;wBAAE,UAAU,CAAC,QAAQ,CAAC,CAAC;gBACpC,CAAC;qBAAM,IAAI,QAAQ,IAAI,CAAC,KAAK,IAAI,CAAC,cAAc,IAAI,KAAK,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC;oBAClE,OAAO,IAAI,CAAC,IAAI,CAAC,EAAE,MAAM,EAAE,IAAI,EAAE,MAAM,EAAE,CAAC,CAAC;gBAC5C,CAAC;gBACD,MAAM,KAAK,CAAC,GAAG,EAAE,MAAM,CAAC,CAAC;YAC1B,CAAC;QACF,CAAC;gBAAS,CAAC;YACV,MAAM,EAAE,CAAC;YACT,OAAO,CAAC,KAAK,EAAE,CAAC;QACjB,CAAC;IAAA,CACD;CACD","sourcesContent":["export type LlamaModelStatus = \"unloaded\" | \"loading\" | \"loaded\" | \"downloading\" | \"sleeping\";\n\nexport interface LlamaModelInfo {\n\tid: string;\n\taliases?: string[];\n\tstatus: {\n\t\tvalue: LlamaModelStatus;\n\t\targs?: string[];\n\t\tfailed?: boolean;\n\t\texit_code?: number;\n\t\tprogress?: Record<string, { done: number; total: number }>;\n\t};\n\tarchitecture?: {\n\t\tinput_modalities?: string[];\n\t\toutput_modalities?: string[];\n\t};\n\tsource?: string;\n\tmeta?: {\n\t\tn_ctx?: number;\n\t\tn_ctx_train?: number;\n\t\tsize?: number;\n\t\tftype?: string;\n\t};\n}\n\nexport interface LlamaModelsResponse {\n\tdata: LlamaModelInfo[];\n\tobject?: string;\n}\n\nexport interface LlamaModelEvent {\n\tmodel: string;\n\tevent: string;\n\tdata?: unknown;\n}\n\nexport interface LlamaProgress {\n\tmessage: string;\n\tratio?: number;\n\tdetail?: string;\n}\n\nfunction errorMessage(payload: unknown, fallback: string): string {\n\tif (typeof payload !== \"object\" || payload === null) return fallback;\n\tconst error = (payload as { error?: unknown }).error;\n\tif (typeof error !== \"object\" || error === null) return fallback;\n\tconst message = (error as { message?: unknown }).message;\n\treturn typeof message === \"string\" && message ? message : fallback;\n}\n\nfunction isModelInfo(value: unknown): value is LlamaModelInfo {\n\tif (typeof value !== \"object\" || value === null) return false;\n\tconst candidate = value as { id?: unknown; status?: { value?: unknown } };\n\treturn typeof candidate.id === \"string\" && typeof candidate.status?.value === \"string\";\n}\n\nfunction linkSignal(source: AbortSignal | undefined, target: AbortController): () => void {\n\tif (!source) return () => {};\n\tif (source.aborted) {\n\t\ttarget.abort(source.reason);\n\t\treturn () => {};\n\t}\n\tconst abort = () => target.abort(source.reason);\n\tsource.addEventListener(\"abort\", abort, { once: true });\n\treturn () => source.removeEventListener(\"abort\", abort);\n}\n\nfunction sleep(ms: number, signal?: AbortSignal): Promise<void> {\n\treturn new Promise((resolve, reject) => {\n\t\tif (signal?.aborted) {\n\t\t\treject(signal.reason ?? new Error(\"Cancelled\"));\n\t\t\treturn;\n\t\t}\n\t\tconst abort = () => {\n\t\t\tclearTimeout(timeout);\n\t\t\treject(signal?.reason ?? new Error(\"Cancelled\"));\n\t\t};\n\t\tconst timeout = setTimeout(() => {\n\t\t\tsignal?.removeEventListener(\"abort\", abort);\n\t\t\tresolve();\n\t\t}, ms);\n\t\tsignal?.addEventListener(\"abort\", abort, { once: true });\n\t});\n}\n\nfunction parseLoadProgress(data: unknown): LlamaProgress | undefined {\n\tif (typeof data !== \"object\" || data === null) return undefined;\n\tconst progress = (data as { progress?: unknown }).progress;\n\tif (typeof progress !== \"object\" || progress === null) return undefined;\n\tconst value = progress as { stages?: unknown; current?: unknown; stage?: unknown; value?: unknown };\n\tconst stage =\n\t\ttypeof value.current === \"string\" ? value.current : typeof value.stage === \"string\" ? value.stage : undefined;\n\tconst stages = Array.isArray(value.stages)\n\t\t? value.stages.filter((entry): entry is string => typeof entry === \"string\")\n\t\t: [];\n\tconst stageRatio = typeof value.value === \"number\" ? Math.max(0, Math.min(1, value.value)) : undefined;\n\tlet ratio = stageRatio;\n\tif (stage && stages.length > 0) {\n\t\tconst index = stages.indexOf(stage);\n\t\tif (index >= 0) ratio = (index + (stageRatio ?? 0)) / stages.length;\n\t}\n\treturn {\n\t\tmessage: stage ? `Loading ${stage.replaceAll(\"_\", \" \")}` : \"Loading model\",\n\t\tratio,\n\t};\n}\n\nfunction parseDownloadProgress(data: unknown): LlamaProgress | undefined {\n\tif (typeof data !== \"object\" || data === null) return undefined;\n\tconst nested = (data as { progress?: unknown }).progress;\n\tconst files = typeof nested === \"object\" && nested !== null ? nested : data;\n\tlet done = 0;\n\tlet total = 0;\n\tfor (const value of Object.values(files as Record<string, unknown>)) {\n\t\tif (typeof value !== \"object\" || value === null) continue;\n\t\tconst entry = value as { done?: unknown; total?: unknown };\n\t\tif (typeof entry.done !== \"number\" || typeof entry.total !== \"number\") continue;\n\t\tdone += entry.done;\n\t\ttotal += entry.total;\n\t}\n\tif (total <= 0) return undefined;\n\treturn {\n\t\tmessage: \"Downloading model\",\n\t\tratio: done / total,\n\t\tdetail: `${formatBytes(done)} / ${formatBytes(total)}`,\n\t};\n}\n\nexport function formatBytes(bytes: number): string {\n\tif (bytes < 1024) return `${bytes} B`;\n\tconst units = [\"KiB\", \"MiB\", \"GiB\", \"TiB\"];\n\tlet value = bytes / 1024;\n\tlet unit = units[0]!;\n\tfor (let index = 1; index < units.length && value >= 1024; index++) {\n\t\tvalue /= 1024;\n\t\tunit = units[index]!;\n\t}\n\treturn `${value >= 10 ? value.toFixed(1) : value.toFixed(2)} ${unit}`;\n}\n\nexport function normalizeLlamaServerUrl(value: string): string {\n\tconst url = new URL(value.trim());\n\tif (url.protocol !== \"http:\" && url.protocol !== \"https:\") {\n\t\tthrow new Error(\"Server URL must use http or https\");\n\t}\n\turl.hash = \"\";\n\turl.search = \"\";\n\turl.pathname = url.pathname.replace(/\\/+$/u, \"\").replace(/\\/v1$/u, \"\") || \"/\";\n\treturn url.toString().replace(/\\/$/u, \"\");\n}\n\nexport function llamaInferenceUrl(serverUrl: string): string {\n\treturn `${normalizeLlamaServerUrl(serverUrl)}/v1`;\n}\n\nexport class LlamaClient {\n\treadonly serverUrl: string;\n\tprivate readonly apiKey: string | undefined;\n\n\tconstructor(serverUrl: string, apiKey?: string) {\n\t\tthis.serverUrl = normalizeLlamaServerUrl(serverUrl);\n\t\tthis.apiKey = apiKey;\n\t}\n\n\tprivate async request(path: string, init: RequestInit = {}): Promise<unknown> {\n\t\tconst headers = new Headers(init.headers);\n\t\tif (init.body !== undefined) headers.set(\"Content-Type\", \"application/json\");\n\t\tif (this.apiKey) headers.set(\"Authorization\", `Bearer ${this.apiKey}`);\n\t\tconst timeout = AbortSignal.timeout(15_000);\n\t\tconst signal = init.signal ? AbortSignal.any([init.signal, timeout]) : timeout;\n\t\tconst response = await fetch(`${this.serverUrl}${path}`, { ...init, headers, signal });\n\t\tlet payload: unknown;\n\t\ttry {\n\t\t\tpayload = await response.json();\n\t\t} catch {\n\t\t\tpayload = undefined;\n\t\t}\n\t\tif (!response.ok) throw new Error(errorMessage(payload, `llama.cpp returned HTTP ${response.status}`));\n\t\treturn payload;\n\t}\n\n\tasync list(options: { reload?: boolean; signal?: AbortSignal } = {}): Promise<LlamaModelInfo[]> {\n\t\tconst payload = await this.request(`/models${options.reload ? \"?reload=1\" : \"\"}`, { signal: options.signal });\n\t\tif (typeof payload !== \"object\" || payload === null || !Array.isArray((payload as { data?: unknown }).data)) {\n\t\t\tthrow new Error(\"llama.cpp returned an invalid model catalog\");\n\t\t}\n\t\tconst data = (payload as { data: unknown[] }).data;\n\t\tif (!data.every(isModelInfo)) throw new Error(\"Server is not running in llama.cpp router mode\");\n\t\treturn data;\n\t}\n\n\tasync load(model: string, signal?: AbortSignal): Promise<void> {\n\t\tawait this.request(\"/models/load\", { method: \"POST\", body: JSON.stringify({ model }), signal });\n\t}\n\n\tasync unload(model: string, signal?: AbortSignal): Promise<void> {\n\t\tawait this.request(\"/models/unload\", { method: \"POST\", body: JSON.stringify({ model }), signal });\n\t}\n\n\tasync unloadAndWait(model: string, signal?: AbortSignal): Promise<void> {\n\t\tawait this.unload(model, signal);\n\t\twhile (true) {\n\t\t\tconst entry = (await this.list({ signal })).find((candidate) => candidate.id === model);\n\t\t\tif (!entry || entry.status.value === \"unloaded\") return;\n\t\t\tawait sleep(100, signal);\n\t\t}\n\t}\n\n\tasync download(model: string, signal?: AbortSignal): Promise<void> {\n\t\tawait this.request(\"/models\", { method: \"POST\", body: JSON.stringify({ model }), signal });\n\t}\n\n\tasync watch(onEvent: (event: LlamaModelEvent) => void, signal?: AbortSignal): Promise<void> {\n\t\tconst headers = new Headers();\n\t\tif (this.apiKey) headers.set(\"Authorization\", `Bearer ${this.apiKey}`);\n\t\tconst response = await fetch(`${this.serverUrl}/models/sse`, { headers, signal });\n\t\tif (!response.ok || !response.body) throw new Error(`llama.cpp SSE returned HTTP ${response.status}`);\n\t\tconst reader = response.body.getReader();\n\t\tconst decoder = new TextDecoder();\n\t\tlet buffer = \"\";\n\t\twhile (true) {\n\t\t\tconst chunk = await reader.read();\n\t\t\tif (chunk.done) break;\n\t\t\tbuffer += decoder.decode(chunk.value, { stream: true }).replaceAll(\"\\r\\n\", \"\\n\");\n\t\t\tlet boundary = buffer.indexOf(\"\\n\\n\");\n\t\t\twhile (boundary >= 0) {\n\t\t\t\tconst frame = buffer.slice(0, boundary);\n\t\t\t\tbuffer = buffer.slice(boundary + 2);\n\t\t\t\tconst data = frame\n\t\t\t\t\t.split(\"\\n\")\n\t\t\t\t\t.filter((line) => line.startsWith(\"data:\"))\n\t\t\t\t\t.map((line) => line.slice(5).trimStart())\n\t\t\t\t\t.join(\"\\n\");\n\t\t\t\tif (data) {\n\t\t\t\t\ttry {\n\t\t\t\t\t\tconst event = JSON.parse(data) as LlamaModelEvent;\n\t\t\t\t\t\tif (event && typeof event.model === \"string\" && typeof event.event === \"string\") onEvent(event);\n\t\t\t\t\t} catch {\n\t\t\t\t\t\t// Ignore malformed events; catalog polling remains authoritative.\n\t\t\t\t\t}\n\t\t\t\t}\n\t\t\t\tboundary = buffer.indexOf(\"\\n\\n\");\n\t\t\t}\n\t\t}\n\t}\n\n\tasync loadAndWait(\n\t\tmodel: string,\n\t\tonProgress: (progress: LlamaProgress) => void,\n\t\tsignal?: AbortSignal,\n\t): Promise<LlamaModelInfo> {\n\t\tconst watcher = new AbortController();\n\t\tconst unlink = linkSignal(signal, watcher);\n\t\tlet eventLoaded = false;\n\t\tlet eventError: string | undefined;\n\t\tvoid this.watch((event) => {\n\t\t\tif (event.model !== model) return;\n\t\t\tif (event.event !== \"model_status\" && event.event !== \"status_change\") return;\n\t\t\tconst data = event.data as { status?: unknown } | undefined;\n\t\t\tif (data?.status === \"loaded\") eventLoaded = true;\n\t\t\tif (data?.status === \"unloaded\") eventError = \"Model failed to load\";\n\t\t\tconst progress = parseLoadProgress(event.data);\n\t\t\tif (progress) onProgress(progress);\n\t\t}, watcher.signal).catch(() => {});\n\t\ttry {\n\t\t\tawait this.load(model, signal);\n\t\t\tonProgress({ message: \"Loading model\" });\n\t\t\twhile (true) {\n\t\t\t\tif (signal?.aborted) throw signal.reason ?? new Error(\"Cancelled\");\n\t\t\t\tconst entry = (await this.list({ signal })).find((candidate) => candidate.id === model);\n\t\t\t\tif (entry?.status.value === \"loaded\") return entry;\n\t\t\t\tif (eventLoaded && !entry) return { id: model, status: { value: \"loaded\" } };\n\t\t\t\tif (entry?.status.failed || eventError) {\n\t\t\t\t\tthrow new Error(\n\t\t\t\t\t\tentry?.status.exit_code === undefined\n\t\t\t\t\t\t\t? (eventError ?? \"Model failed to load\")\n\t\t\t\t\t\t\t: `Model exited with code ${entry.status.exit_code}`,\n\t\t\t\t\t);\n\t\t\t\t}\n\t\t\t\tawait sleep(250, signal);\n\t\t\t}\n\t\t} finally {\n\t\t\tunlink();\n\t\t\twatcher.abort();\n\t\t}\n\t}\n\n\tasync downloadAndWait(\n\t\tmodel: string,\n\t\tonProgress: (progress: LlamaProgress) => void,\n\t\tsignal?: AbortSignal,\n\t): Promise<LlamaModelInfo[]> {\n\t\tconst watcher = new AbortController();\n\t\tconst unlink = linkSignal(signal, watcher);\n\t\tlet finished = false;\n\t\tlet failure: string | undefined;\n\t\tlet sawDownloading = false;\n\t\tlet polls = 0;\n\t\tvoid this.watch((event) => {\n\t\t\tif (event.model !== model) return;\n\t\t\tif (event.event === \"download_finished\") finished = true;\n\t\t\tif (event.event === \"download_failed\") failure = errorMessage(event.data, \"Download failed\");\n\t\t\tif (event.event === \"download_progress\") {\n\t\t\t\tsawDownloading = true;\n\t\t\t\tconst progress = parseDownloadProgress(event.data);\n\t\t\t\tif (progress) onProgress(progress);\n\t\t\t}\n\t\t}, watcher.signal).catch(() => {});\n\t\ttry {\n\t\t\tawait this.download(model, signal);\n\t\t\tonProgress({ message: \"Downloading model\" });\n\t\t\twhile (true) {\n\t\t\t\tif (signal?.aborted) throw signal.reason ?? new Error(\"Cancelled\");\n\t\t\t\tif (failure) throw new Error(failure);\n\t\t\t\tconst models = await this.list({ signal });\n\t\t\t\tpolls++;\n\t\t\t\tconst entry = models.find((candidate) => candidate.id === model);\n\t\t\t\tif (entry?.status.value === \"downloading\") {\n\t\t\t\t\tsawDownloading = true;\n\t\t\t\t\tconst progress = parseDownloadProgress(entry.status.progress);\n\t\t\t\t\tif (progress) onProgress(progress);\n\t\t\t\t} else if (finished || (entry && (sawDownloading || polls >= 2))) {\n\t\t\t\t\treturn this.list({ reload: true, signal });\n\t\t\t\t}\n\t\t\t\tawait sleep(500, signal);\n\t\t\t}\n\t\t} finally {\n\t\t\tunlink();\n\t\t\twatcher.abort();\n\t\t}\n\t}\n}\n"]}
@@ -1,23 +0,0 @@
1
- export interface HuggingFaceModel {
2
- id: string;
3
- downloads: number;
4
- }
5
- export interface HuggingFaceQuantization {
6
- name: string;
7
- size?: number;
8
- }
9
- export interface HuggingFaceModelDetails {
10
- id: string;
11
- gated: false | "auto" | "manual";
12
- quantizations: HuggingFaceQuantization[];
13
- }
14
- export declare function findHuggingFaceToken(env?: NodeJS.ProcessEnv): Promise<string | undefined>;
15
- export declare class HuggingFaceClient {
16
- private readonly token;
17
- private readonly baseUrl;
18
- constructor(token?: string, baseUrl?: string);
19
- private request;
20
- search(query: string, signal?: AbortSignal): Promise<HuggingFaceModel[]>;
21
- details(id: string, signal?: AbortSignal): Promise<HuggingFaceModelDetails>;
22
- }
23
- //# sourceMappingURL=huggingface.d.ts.map
@@ -1 +0,0 @@
1
- {"version":3,"file":"huggingface.d.ts","sourceRoot":"","sources":["../../../src/extensions/llama/huggingface.ts"],"names":[],"mappings":"AASA,MAAM,WAAW,gBAAgB;IAChC,EAAE,EAAE,MAAM,CAAC;IACX,SAAS,EAAE,MAAM,CAAC;CAClB;AAED,MAAM,WAAW,uBAAuB;IACvC,IAAI,EAAE,MAAM,CAAC;IACb,IAAI,CAAC,EAAE,MAAM,CAAC;CACd;AAED,MAAM,WAAW,uBAAuB;IACvC,EAAE,EAAE,MAAM,CAAC;IACX,KAAK,EAAE,KAAK,GAAG,MAAM,GAAG,QAAQ,CAAC;IACjC,aAAa,EAAE,uBAAuB,EAAE,CAAC;CACzC;AAsBD,wBAAsB,oBAAoB,CAAC,GAAG,GAAE,MAAM,CAAC,UAAwB,GAAG,OAAO,CAAC,MAAM,GAAG,SAAS,CAAC,CAe5G;AAED,qBAAa,iBAAiB;IAC7B,OAAO,CAAC,QAAQ,CAAC,KAAK,CAAqB;IAC3C,OAAO,CAAC,QAAQ,CAAC,OAAO,CAAS;IAEjC,YAAY,KAAK,CAAC,EAAE,MAAM,EAAE,OAAO,SAA2B,EAG7D;YAEa,OAAO;IA4Bf,MAAM,CAAC,KAAK,EAAE,MAAM,EAAE,MAAM,CAAC,EAAE,WAAW,GAAG,OAAO,CAAC,gBAAgB,EAAE,CAAC,CAgB7E;IAEK,OAAO,CAAC,EAAE,EAAE,MAAM,EAAE,MAAM,CAAC,EAAE,WAAW,GAAG,OAAO,CAAC,uBAAuB,CAAC,CAuChF;CACD","sourcesContent":["import { readFile } from \"node:fs/promises\";\nimport { homedir } from \"node:os\";\nimport { join } from \"node:path\";\n\nconst DEFAULT_HUGGING_FACE_URL = \"https://huggingface.co\";\nconst QUANTIZATION_PATTERN =\n\t/(?:^|[-_.])((?:UD-)?(?:IQ\\d(?:_[A-Z0-9]+)+|Q\\d(?:_[A-Z0-9]+)+|BF16|F16|F32|MXFP\\d(?:_[A-Z0-9]+)*))$/iu;\nconst SHARD_SUFFIX_PATTERN = /-\\d{5}-of-\\d{5}$/u;\n\nexport interface HuggingFaceModel {\n\tid: string;\n\tdownloads: number;\n}\n\nexport interface HuggingFaceQuantization {\n\tname: string;\n\tsize?: number;\n}\n\nexport interface HuggingFaceModelDetails {\n\tid: string;\n\tgated: false | \"auto\" | \"manual\";\n\tquantizations: HuggingFaceQuantization[];\n}\n\nfunction payloadError(payload: unknown, fallback: string): string {\n\tif (typeof payload !== \"object\" || payload === null) return fallback;\n\tconst error = (payload as { error?: unknown }).error;\n\treturn typeof error === \"string\" && error ? error : fallback;\n}\n\nfunction parseRateLimitDelay(value: string | null): number | undefined {\n\tconst match = value?.match(/(?:^|;)t=(\\d+)/u);\n\treturn match ? Number(match[1]) : undefined;\n}\n\nasync function readToken(path: string): Promise<string | undefined> {\n\ttry {\n\t\tconst token = (await readFile(path, \"utf8\")).trim();\n\t\treturn token || undefined;\n\t} catch {\n\t\treturn undefined;\n\t}\n}\n\nexport async function findHuggingFaceToken(env: NodeJS.ProcessEnv = process.env): Promise<string | undefined> {\n\tconst fromEnvironment = env.HF_TOKEN?.trim();\n\tif (fromEnvironment) return fromEnvironment;\n\n\tconst paths = [\n\t\tenv.HF_TOKEN_PATH,\n\t\tenv.HF_HOME ? join(env.HF_HOME, \"token\") : undefined,\n\t\tenv.XDG_CACHE_HOME ? join(env.XDG_CACHE_HOME, \"huggingface\", \"token\") : undefined,\n\t\tjoin(homedir(), \".cache\", \"huggingface\", \"token\"),\n\t].filter((path): path is string => Boolean(path));\n\tfor (const path of new Set(paths)) {\n\t\tconst token = await readToken(path);\n\t\tif (token) return token;\n\t}\n\treturn undefined;\n}\n\nexport class HuggingFaceClient {\n\tprivate readonly token: string | undefined;\n\tprivate readonly baseUrl: string;\n\n\tconstructor(token?: string, baseUrl = DEFAULT_HUGGING_FACE_URL) {\n\t\tthis.token = token;\n\t\tthis.baseUrl = baseUrl.replace(/\\/+$/u, \"\");\n\t}\n\n\tprivate async request(path: string, signal?: AbortSignal): Promise<unknown> {\n\t\tconst headers = new Headers();\n\t\tif (this.token) headers.set(\"Authorization\", `Bearer ${this.token}`);\n\t\tconst timeout = AbortSignal.timeout(15_000);\n\t\tconst response = await fetch(`${this.baseUrl}${path}`, {\n\t\t\theaders,\n\t\t\tsignal: signal ? AbortSignal.any([signal, timeout]) : timeout,\n\t\t});\n\t\tlet payload: unknown;\n\t\ttry {\n\t\t\tpayload = await response.json();\n\t\t} catch {\n\t\t\tpayload = undefined;\n\t\t}\n\t\tif (!response.ok) {\n\t\t\tconst fallback = `Hugging Face returned HTTP ${response.status}`;\n\t\t\tif (response.status === 429) {\n\t\t\t\tconst delay =\n\t\t\t\t\tNumber(response.headers.get(\"retry-after\")) || parseRateLimitDelay(response.headers.get(\"ratelimit\"));\n\t\t\t\tthrow new Error(\n\t\t\t\t\tdelay ? `Hugging Face rate limit reached; retry in ${delay}s` : \"Hugging Face rate limit reached\",\n\t\t\t\t);\n\t\t\t}\n\t\t\tthrow new Error(payloadError(payload, fallback));\n\t\t}\n\t\treturn payload;\n\t}\n\n\tasync search(query: string, signal?: AbortSignal): Promise<HuggingFaceModel[]> {\n\t\tconst params = new URLSearchParams({\n\t\t\tsearch: query,\n\t\t\tfilter: \"gguf\",\n\t\t\tsort: \"downloads\",\n\t\t\tdirection: \"-1\",\n\t\t\tlimit: \"20\",\n\t\t});\n\t\tconst payload = await this.request(`/api/models?${params}`, signal);\n\t\tif (!Array.isArray(payload)) throw new Error(\"Hugging Face returned invalid search results\");\n\t\treturn payload.flatMap((value) => {\n\t\t\tif (typeof value !== \"object\" || value === null || typeof (value as { id?: unknown }).id !== \"string\")\n\t\t\t\treturn [];\n\t\t\tconst model = value as { id: string; downloads?: unknown };\n\t\t\treturn [{ id: model.id, downloads: typeof model.downloads === \"number\" ? model.downloads : 0 }];\n\t\t});\n\t}\n\n\tasync details(id: string, signal?: AbortSignal): Promise<HuggingFaceModelDetails> {\n\t\tconst encodedId = id.split(\"/\").map(encodeURIComponent).join(\"/\");\n\t\tconst payload = await this.request(`/api/models/${encodedId}?blobs=true`, signal);\n\t\tif (typeof payload !== \"object\" || payload === null) {\n\t\t\tthrow new Error(\"Hugging Face returned invalid model details\");\n\t\t}\n\t\tconst model = payload as { id?: unknown; gated?: unknown; siblings?: unknown };\n\t\tconst sizes = new Map<string, { total: number; complete: boolean }>();\n\t\tif (Array.isArray(model.siblings)) {\n\t\t\tfor (const value of model.siblings) {\n\t\t\t\tif (typeof value !== \"object\" || value === null) continue;\n\t\t\t\tconst file = value as { rfilename?: unknown; size?: unknown };\n\t\t\t\tif (typeof file.rfilename !== \"string\" || !file.rfilename.toLowerCase().endsWith(\".gguf\")) continue;\n\t\t\t\tconst filename = file.rfilename.split(\"/\").at(-1)!;\n\t\t\t\tif (filename.toLowerCase().startsWith(\"mmproj\")) continue;\n\t\t\t\tconst stem = filename.slice(0, -5).replace(SHARD_SUFFIX_PATTERN, \"\");\n\t\t\t\tconst quantization = stem.match(QUANTIZATION_PATTERN)?.[1]?.toUpperCase();\n\t\t\t\tif (!quantization) continue;\n\t\t\t\tconst current = sizes.get(quantization) ?? { total: 0, complete: true };\n\t\t\t\tif (typeof file.size === \"number\") current.total += file.size;\n\t\t\t\telse current.complete = false;\n\t\t\t\tsizes.set(quantization, current);\n\t\t\t}\n\t\t}\n\t\tconst quantizations = [...sizes]\n\t\t\t.map(([name, size]) => ({ name, size: size.complete ? size.total : undefined }))\n\t\t\t.sort((left, right) => {\n\t\t\t\tif (left.name === \"Q4_K_M\") return -1;\n\t\t\t\tif (right.name === \"Q4_K_M\") return 1;\n\t\t\t\treturn (\n\t\t\t\t\t(left.size ?? Number.MAX_SAFE_INTEGER) - (right.size ?? Number.MAX_SAFE_INTEGER) ||\n\t\t\t\t\tleft.name.localeCompare(right.name)\n\t\t\t\t);\n\t\t\t});\n\t\treturn {\n\t\t\tid: typeof model.id === \"string\" ? model.id : id,\n\t\t\tgated: model.gated === \"auto\" || model.gated === \"manual\" ? model.gated : false,\n\t\t\tquantizations,\n\t\t};\n\t}\n}\n"]}
@@ -1,146 +0,0 @@
1
- "use strict";
2
- Object.defineProperty(exports, "__esModule", { value: true });
3
- exports.HuggingFaceClient = void 0;
4
- exports.findHuggingFaceToken = findHuggingFaceToken;
5
- const promises_1 = require("node:fs/promises");
6
- const node_os_1 = require("node:os");
7
- const node_path_1 = require("node:path");
8
- const DEFAULT_HUGGING_FACE_URL = "https://huggingface.co";
9
- const QUANTIZATION_PATTERN = /(?:^|[-_.])((?:UD-)?(?:IQ\d(?:_[A-Z0-9]+)+|Q\d(?:_[A-Z0-9]+)+|BF16|F16|F32|MXFP\d(?:_[A-Z0-9]+)*))$/iu;
10
- const SHARD_SUFFIX_PATTERN = /-\d{5}-of-\d{5}$/u;
11
- function payloadError(payload, fallback) {
12
- if (typeof payload !== "object" || payload === null)
13
- return fallback;
14
- const error = payload.error;
15
- return typeof error === "string" && error ? error : fallback;
16
- }
17
- function parseRateLimitDelay(value) {
18
- const match = value?.match(/(?:^|;)t=(\d+)/u);
19
- return match ? Number(match[1]) : undefined;
20
- }
21
- async function readToken(path) {
22
- try {
23
- const token = (await (0, promises_1.readFile)(path, "utf8")).trim();
24
- return token || undefined;
25
- }
26
- catch {
27
- return undefined;
28
- }
29
- }
30
- async function findHuggingFaceToken(env = process.env) {
31
- const fromEnvironment = env.HF_TOKEN?.trim();
32
- if (fromEnvironment)
33
- return fromEnvironment;
34
- const paths = [
35
- env.HF_TOKEN_PATH,
36
- env.HF_HOME ? (0, node_path_1.join)(env.HF_HOME, "token") : undefined,
37
- env.XDG_CACHE_HOME ? (0, node_path_1.join)(env.XDG_CACHE_HOME, "huggingface", "token") : undefined,
38
- (0, node_path_1.join)((0, node_os_1.homedir)(), ".cache", "huggingface", "token"),
39
- ].filter((path) => Boolean(path));
40
- for (const path of new Set(paths)) {
41
- const token = await readToken(path);
42
- if (token)
43
- return token;
44
- }
45
- return undefined;
46
- }
47
- class HuggingFaceClient {
48
- token;
49
- baseUrl;
50
- constructor(token, baseUrl = DEFAULT_HUGGING_FACE_URL) {
51
- this.token = token;
52
- this.baseUrl = baseUrl.replace(/\/+$/u, "");
53
- }
54
- async request(path, signal) {
55
- const headers = new Headers();
56
- if (this.token)
57
- headers.set("Authorization", `Bearer ${this.token}`);
58
- const timeout = AbortSignal.timeout(15_000);
59
- const response = await fetch(`${this.baseUrl}${path}`, {
60
- headers,
61
- signal: signal ? AbortSignal.any([signal, timeout]) : timeout,
62
- });
63
- let payload;
64
- try {
65
- payload = await response.json();
66
- }
67
- catch {
68
- payload = undefined;
69
- }
70
- if (!response.ok) {
71
- const fallback = `Hugging Face returned HTTP ${response.status}`;
72
- if (response.status === 429) {
73
- const delay = Number(response.headers.get("retry-after")) || parseRateLimitDelay(response.headers.get("ratelimit"));
74
- throw new Error(delay ? `Hugging Face rate limit reached; retry in ${delay}s` : "Hugging Face rate limit reached");
75
- }
76
- throw new Error(payloadError(payload, fallback));
77
- }
78
- return payload;
79
- }
80
- async search(query, signal) {
81
- const params = new URLSearchParams({
82
- search: query,
83
- filter: "gguf",
84
- sort: "downloads",
85
- direction: "-1",
86
- limit: "20",
87
- });
88
- const payload = await this.request(`/api/models?${params}`, signal);
89
- if (!Array.isArray(payload))
90
- throw new Error("Hugging Face returned invalid search results");
91
- return payload.flatMap((value) => {
92
- if (typeof value !== "object" || value === null || typeof value.id !== "string")
93
- return [];
94
- const model = value;
95
- return [{ id: model.id, downloads: typeof model.downloads === "number" ? model.downloads : 0 }];
96
- });
97
- }
98
- async details(id, signal) {
99
- const encodedId = id.split("/").map(encodeURIComponent).join("/");
100
- const payload = await this.request(`/api/models/${encodedId}?blobs=true`, signal);
101
- if (typeof payload !== "object" || payload === null) {
102
- throw new Error("Hugging Face returned invalid model details");
103
- }
104
- const model = payload;
105
- const sizes = new Map();
106
- if (Array.isArray(model.siblings)) {
107
- for (const value of model.siblings) {
108
- if (typeof value !== "object" || value === null)
109
- continue;
110
- const file = value;
111
- if (typeof file.rfilename !== "string" || !file.rfilename.toLowerCase().endsWith(".gguf"))
112
- continue;
113
- const filename = file.rfilename.split("/").at(-1);
114
- if (filename.toLowerCase().startsWith("mmproj"))
115
- continue;
116
- const stem = filename.slice(0, -5).replace(SHARD_SUFFIX_PATTERN, "");
117
- const quantization = stem.match(QUANTIZATION_PATTERN)?.[1]?.toUpperCase();
118
- if (!quantization)
119
- continue;
120
- const current = sizes.get(quantization) ?? { total: 0, complete: true };
121
- if (typeof file.size === "number")
122
- current.total += file.size;
123
- else
124
- current.complete = false;
125
- sizes.set(quantization, current);
126
- }
127
- }
128
- const quantizations = [...sizes]
129
- .map(([name, size]) => ({ name, size: size.complete ? size.total : undefined }))
130
- .sort((left, right) => {
131
- if (left.name === "Q4_K_M")
132
- return -1;
133
- if (right.name === "Q4_K_M")
134
- return 1;
135
- return ((left.size ?? Number.MAX_SAFE_INTEGER) - (right.size ?? Number.MAX_SAFE_INTEGER) ||
136
- left.name.localeCompare(right.name));
137
- });
138
- return {
139
- id: typeof model.id === "string" ? model.id : id,
140
- gated: model.gated === "auto" || model.gated === "manual" ? model.gated : false,
141
- quantizations,
142
- };
143
- }
144
- }
145
- exports.HuggingFaceClient = HuggingFaceClient;
146
- //# sourceMappingURL=huggingface.js.map
@@ -1 +0,0 @@
1
- {"version":3,"file":"huggingface.js","sourceRoot":"","sources":["../../../src/extensions/llama/huggingface.ts"],"names":[],"mappings":";;;;AAAA,+CAA4C;AAC5C,qCAAkC;AAClC,yCAAiC;AAEjC,MAAM,wBAAwB,GAAG,wBAAwB,CAAC;AAC1D,MAAM,oBAAoB,GACzB,uGAAuG,CAAC;AACzG,MAAM,oBAAoB,GAAG,mBAAmB,CAAC;AAkBjD,SAAS,YAAY,CAAC,OAAgB,EAAE,QAAgB,EAAU;IACjE,IAAI,OAAO,OAAO,KAAK,QAAQ,IAAI,OAAO,KAAK,IAAI;QAAE,OAAO,QAAQ,CAAC;IACrE,MAAM,KAAK,GAAI,OAA+B,CAAC,KAAK,CAAC;IACrD,OAAO,OAAO,KAAK,KAAK,QAAQ,IAAI,KAAK,CAAC,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,QAAQ,CAAC;AAAA,CAC7D;AAED,SAAS,mBAAmB,CAAC,KAAoB,EAAsB;IACtE,MAAM,KAAK,GAAG,KAAK,EAAE,KAAK,CAAC,iBAAiB,CAAC,CAAC;IAC9C,OAAO,KAAK,CAAC,CAAC,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC;AAAA,CAC5C;AAED,KAAK,UAAU,SAAS,CAAC,IAAY,EAA+B;IACnE,IAAI,CAAC;QACJ,MAAM,KAAK,GAAG,CAAC,MAAM,IAAA,mBAAQ,EAAC,IAAI,EAAE,MAAM,CAAC,CAAC,CAAC,IAAI,EAAE,CAAC;QACpD,OAAO,KAAK,IAAI,SAAS,CAAC;IAC3B,CAAC;IAAC,MAAM,CAAC;QACR,OAAO,SAAS,CAAC;IAClB,CAAC;AAAA,CACD;AAEM,KAAK,+BAA+B,GAAG,GAAsB,OAAO,CAAC,GAAG,EAA+B;IAC7G,MAAM,eAAe,GAAG,GAAG,CAAC,QAAQ,EAAE,IAAI,EAAE,CAAC;IAC7C,IAAI,eAAe;QAAE,OAAO,eAAe,CAAC;IAE5C,MAAM,KAAK,GAAG;QACb,GAAG,CAAC,aAAa;QACjB,GAAG,CAAC,OAAO,CAAC,CAAC,CAAC,IAAA,gBAAI,EAAC,GAAG,CAAC,OAAO,EAAE,OAAO,CAAC,CAAC,CAAC,CAAC,SAAS;QACpD,GAAG,CAAC,cAAc,CAAC,CAAC,CAAC,IAAA,gBAAI,EAAC,GAAG,CAAC,cAAc,EAAE,aAAa,EAAE,OAAO,CAAC,CAAC,CAAC,CAAC,SAAS;QACjF,IAAA,gBAAI,EAAC,IAAA,iBAAO,GAAE,EAAE,QAAQ,EAAE,aAAa,EAAE,OAAO,CAAC;KACjD,CAAC,MAAM,CAAC,CAAC,IAAI,EAAkB,EAAE,CAAC,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC;IAClD,KAAK,MAAM,IAAI,IAAI,IAAI,GAAG,CAAC,KAAK,CAAC,EAAE,CAAC;QACnC,MAAM,KAAK,GAAG,MAAM,SAAS,CAAC,IAAI,CAAC,CAAC;QACpC,IAAI,KAAK;YAAE,OAAO,KAAK,CAAC;IACzB,CAAC;IACD,OAAO,SAAS,CAAC;AAAA,CACjB;AAED;IACkB,KAAK,CAAqB;IAC1B,OAAO,CAAS;IAEjC,YAAY,KAAc,EAAE,OAAO,GAAG,wBAAwB,EAAE;QAC/D,IAAI,CAAC,KAAK,GAAG,KAAK,CAAC;QACnB,IAAI,CAAC,OAAO,GAAG,OAAO,CAAC,OAAO,CAAC,OAAO,EAAE,EAAE,CAAC,CAAC;IAAA,CAC5C;IAEO,KAAK,CAAC,OAAO,CAAC,IAAY,EAAE,MAAoB,EAAoB;QAC3E,MAAM,OAAO,GAAG,IAAI,OAAO,EAAE,CAAC;QAC9B,IAAI,IAAI,CAAC,KAAK;YAAE,OAAO,CAAC,GAAG,CAAC,eAAe,EAAE,UAAU,IAAI,CAAC,KAAK,EAAE,CAAC,CAAC;QACrE,MAAM,OAAO,GAAG,WAAW,CAAC,OAAO,CAAC,MAAM,CAAC,CAAC;QAC5C,MAAM,QAAQ,GAAG,MAAM,KAAK,CAAC,GAAG,IAAI,CAAC,OAAO,GAAG,IAAI,EAAE,EAAE;YACtD,OAAO;YACP,MAAM,EAAE,MAAM,CAAC,CAAC,CAAC,WAAW,CAAC,GAAG,CAAC,CAAC,MAAM,EAAE,OAAO,CAAC,CAAC,CAAC,CAAC,CAAC,OAAO;SAC7D,CAAC,CAAC;QACH,IAAI,OAAgB,CAAC;QACrB,IAAI,CAAC;YACJ,OAAO,GAAG,MAAM,QAAQ,CAAC,IAAI,EAAE,CAAC;QACjC,CAAC;QAAC,MAAM,CAAC;YACR,OAAO,GAAG,SAAS,CAAC;QACrB,CAAC;QACD,IAAI,CAAC,QAAQ,CAAC,EAAE,EAAE,CAAC;YAClB,MAAM,QAAQ,GAAG,8BAA8B,QAAQ,CAAC,MAAM,EAAE,CAAC;YACjE,IAAI,QAAQ,CAAC,MAAM,KAAK,GAAG,EAAE,CAAC;gBAC7B,MAAM,KAAK,GACV,MAAM,CAAC,QAAQ,CAAC,OAAO,CAAC,GAAG,CAAC,aAAa,CAAC,CAAC,IAAI,mBAAmB,CAAC,QAAQ,CAAC,OAAO,CAAC,GAAG,CAAC,WAAW,CAAC,CAAC,CAAC;gBACvG,MAAM,IAAI,KAAK,CACd,KAAK,CAAC,CAAC,CAAC,6CAA6C,KAAK,GAAG,CAAC,CAAC,CAAC,iCAAiC,CACjG,CAAC;YACH,CAAC;YACD,MAAM,IAAI,KAAK,CAAC,YAAY,CAAC,OAAO,EAAE,QAAQ,CAAC,CAAC,CAAC;QAClD,CAAC;QACD,OAAO,OAAO,CAAC;IAAA,CACf;IAED,KAAK,CAAC,MAAM,CAAC,KAAa,EAAE,MAAoB,EAA+B;QAC9E,MAAM,MAAM,GAAG,IAAI,eAAe,CAAC;YAClC,MAAM,EAAE,KAAK;YACb,MAAM,EAAE,MAAM;YACd,IAAI,EAAE,WAAW;YACjB,SAAS,EAAE,IAAI;YACf,KAAK,EAAE,IAAI;SACX,CAAC,CAAC;QACH,MAAM,OAAO,GAAG,MAAM,IAAI,CAAC,OAAO,CAAC,eAAe,MAAM,EAAE,EAAE,MAAM,CAAC,CAAC;QACpE,IAAI,CAAC,KAAK,CAAC,OAAO,CAAC,OAAO,CAAC;YAAE,MAAM,IAAI,KAAK,CAAC,8CAA8C,CAAC,CAAC;QAC7F,OAAO,OAAO,CAAC,OAAO,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC;YACjC,IAAI,OAAO,KAAK,KAAK,QAAQ,IAAI,KAAK,KAAK,IAAI,IAAI,OAAQ,KAA0B,CAAC,EAAE,KAAK,QAAQ;gBACpG,OAAO,EAAE,CAAC;YACX,MAAM,KAAK,GAAG,KAA4C,CAAC;YAC3D,OAAO,CAAC,EAAE,EAAE,EAAE,KAAK,CAAC,EAAE,EAAE,SAAS,EAAE,OAAO,KAAK,CAAC,SAAS,KAAK,QAAQ,CAAC,CAAC,CAAC,KAAK,CAAC,SAAS,CAAC,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC;QAAA,CAChG,CAAC,CAAC;IAAA,CACH;IAED,KAAK,CAAC,OAAO,CAAC,EAAU,EAAE,MAAoB,EAAoC;QACjF,MAAM,SAAS,GAAG,EAAE,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC,GAAG,CAAC,kBAAkB,CAAC,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC;QAClE,MAAM,OAAO,GAAG,MAAM,IAAI,CAAC,OAAO,CAAC,eAAe,SAAS,aAAa,EAAE,MAAM,CAAC,CAAC;QAClF,IAAI,OAAO,OAAO,KAAK,QAAQ,IAAI,OAAO,KAAK,IAAI,EAAE,CAAC;YACrD,MAAM,IAAI,KAAK,CAAC,6CAA6C,CAAC,CAAC;QAChE,CAAC;QACD,MAAM,KAAK,GAAG,OAAgE,CAAC;QAC/E,MAAM,KAAK,GAAG,IAAI,GAAG,EAAgD,CAAC;QACtE,IAAI,KAAK,CAAC,OAAO,CAAC,KAAK,CAAC,QAAQ,CAAC,EAAE,CAAC;YACnC,KAAK,MAAM,KAAK,IAAI,KAAK,CAAC,QAAQ,EAAE,CAAC;gBACpC,IAAI,OAAO,KAAK,KAAK,QAAQ,IAAI,KAAK,KAAK,IAAI;oBAAE,SAAS;gBAC1D,MAAM,IAAI,GAAG,KAAgD,CAAC;gBAC9D,IAAI,OAAO,IAAI,CAAC,SAAS,KAAK,QAAQ,IAAI,CAAC,IAAI,CAAC,SAAS,CAAC,WAAW,EAAE,CAAC,QAAQ,CAAC,OAAO,CAAC;oBAAE,SAAS;gBACpG,MAAM,QAAQ,GAAG,IAAI,CAAC,SAAS,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,CAAE,CAAC;gBACnD,IAAI,QAAQ,CAAC,WAAW,EAAE,CAAC,UAAU,CAAC,QAAQ,CAAC;oBAAE,SAAS;gBAC1D,MAAM,IAAI,GAAG,QAAQ,CAAC,KAAK,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,CAAC,OAAO,CAAC,oBAAoB,EAAE,EAAE,CAAC,CAAC;gBACrE,MAAM,YAAY,GAAG,IAAI,CAAC,KAAK,CAAC,oBAAoB,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE,WAAW,EAAE,CAAC;gBAC1E,IAAI,CAAC,YAAY;oBAAE,SAAS;gBAC5B,MAAM,OAAO,GAAG,KAAK,CAAC,GAAG,CAAC,YAAY,CAAC,IAAI,EAAE,KAAK,EAAE,CAAC,EAAE,QAAQ,EAAE,IAAI,EAAE,CAAC;gBACxE,IAAI,OAAO,IAAI,CAAC,IAAI,KAAK,QAAQ;oBAAE,OAAO,CAAC,KAAK,IAAI,IAAI,CAAC,IAAI,CAAC;;oBACzD,OAAO,CAAC,QAAQ,GAAG,KAAK,CAAC;gBAC9B,KAAK,CAAC,GAAG,CAAC,YAAY,EAAE,OAAO,CAAC,CAAC;YAClC,CAAC;QACF,CAAC;QACD,MAAM,aAAa,GAAG,CAAC,GAAG,KAAK,CAAC;aAC9B,GAAG,CAAC,CAAC,CAAC,IAAI,EAAE,IAAI,CAAC,EAAE,EAAE,CAAC,CAAC,EAAE,IAAI,EAAE,IAAI,EAAE,IAAI,CAAC,QAAQ,CAAC,CAAC,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC,CAAC,SAAS,EAAE,CAAC,CAAC;aAC/E,IAAI,CAAC,CAAC,IAAI,EAAE,KAAK,EAAE,EAAE,CAAC;YACtB,IAAI,IAAI,CAAC,IAAI,KAAK,QAAQ;gBAAE,OAAO,CAAC,CAAC,CAAC;YACtC,IAAI,KAAK,CAAC,IAAI,KAAK,QAAQ;gBAAE,OAAO,CAAC,CAAC;YACtC,OAAO,CACN,CAAC,IAAI,CAAC,IAAI,IAAI,MAAM,CAAC,gBAAgB,CAAC,GAAG,CAAC,KAAK,CAAC,IAAI,IAAI,MAAM,CAAC,gBAAgB,CAAC;gBAChF,IAAI,CAAC,IAAI,CAAC,aAAa,CAAC,KAAK,CAAC,IAAI,CAAC,CACnC,CAAC;QAAA,CACF,CAAC,CAAC;QACJ,OAAO;YACN,EAAE,EAAE,OAAO,KAAK,CAAC,EAAE,KAAK,QAAQ,CAAC,CAAC,CAAC,KAAK,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE;YAChD,KAAK,EAAE,KAAK,CAAC,KAAK,KAAK,MAAM,IAAI,KAAK,CAAC,KAAK,KAAK,QAAQ,CAAC,CAAC,CAAC,KAAK,CAAC,KAAK,CAAC,CAAC,CAAC,KAAK;YAC/E,aAAa;SACb,CAAC;IAAA,CACF;CACD","sourcesContent":["import { readFile } from \"node:fs/promises\";\nimport { homedir } from \"node:os\";\nimport { join } from \"node:path\";\n\nconst DEFAULT_HUGGING_FACE_URL = \"https://huggingface.co\";\nconst QUANTIZATION_PATTERN =\n\t/(?:^|[-_.])((?:UD-)?(?:IQ\\d(?:_[A-Z0-9]+)+|Q\\d(?:_[A-Z0-9]+)+|BF16|F16|F32|MXFP\\d(?:_[A-Z0-9]+)*))$/iu;\nconst SHARD_SUFFIX_PATTERN = /-\\d{5}-of-\\d{5}$/u;\n\nexport interface HuggingFaceModel {\n\tid: string;\n\tdownloads: number;\n}\n\nexport interface HuggingFaceQuantization {\n\tname: string;\n\tsize?: number;\n}\n\nexport interface HuggingFaceModelDetails {\n\tid: string;\n\tgated: false | \"auto\" | \"manual\";\n\tquantizations: HuggingFaceQuantization[];\n}\n\nfunction payloadError(payload: unknown, fallback: string): string {\n\tif (typeof payload !== \"object\" || payload === null) return fallback;\n\tconst error = (payload as { error?: unknown }).error;\n\treturn typeof error === \"string\" && error ? error : fallback;\n}\n\nfunction parseRateLimitDelay(value: string | null): number | undefined {\n\tconst match = value?.match(/(?:^|;)t=(\\d+)/u);\n\treturn match ? Number(match[1]) : undefined;\n}\n\nasync function readToken(path: string): Promise<string | undefined> {\n\ttry {\n\t\tconst token = (await readFile(path, \"utf8\")).trim();\n\t\treturn token || undefined;\n\t} catch {\n\t\treturn undefined;\n\t}\n}\n\nexport async function findHuggingFaceToken(env: NodeJS.ProcessEnv = process.env): Promise<string | undefined> {\n\tconst fromEnvironment = env.HF_TOKEN?.trim();\n\tif (fromEnvironment) return fromEnvironment;\n\n\tconst paths = [\n\t\tenv.HF_TOKEN_PATH,\n\t\tenv.HF_HOME ? join(env.HF_HOME, \"token\") : undefined,\n\t\tenv.XDG_CACHE_HOME ? join(env.XDG_CACHE_HOME, \"huggingface\", \"token\") : undefined,\n\t\tjoin(homedir(), \".cache\", \"huggingface\", \"token\"),\n\t].filter((path): path is string => Boolean(path));\n\tfor (const path of new Set(paths)) {\n\t\tconst token = await readToken(path);\n\t\tif (token) return token;\n\t}\n\treturn undefined;\n}\n\nexport class HuggingFaceClient {\n\tprivate readonly token: string | undefined;\n\tprivate readonly baseUrl: string;\n\n\tconstructor(token?: string, baseUrl = DEFAULT_HUGGING_FACE_URL) {\n\t\tthis.token = token;\n\t\tthis.baseUrl = baseUrl.replace(/\\/+$/u, \"\");\n\t}\n\n\tprivate async request(path: string, signal?: AbortSignal): Promise<unknown> {\n\t\tconst headers = new Headers();\n\t\tif (this.token) headers.set(\"Authorization\", `Bearer ${this.token}`);\n\t\tconst timeout = AbortSignal.timeout(15_000);\n\t\tconst response = await fetch(`${this.baseUrl}${path}`, {\n\t\t\theaders,\n\t\t\tsignal: signal ? AbortSignal.any([signal, timeout]) : timeout,\n\t\t});\n\t\tlet payload: unknown;\n\t\ttry {\n\t\t\tpayload = await response.json();\n\t\t} catch {\n\t\t\tpayload = undefined;\n\t\t}\n\t\tif (!response.ok) {\n\t\t\tconst fallback = `Hugging Face returned HTTP ${response.status}`;\n\t\t\tif (response.status === 429) {\n\t\t\t\tconst delay =\n\t\t\t\t\tNumber(response.headers.get(\"retry-after\")) || parseRateLimitDelay(response.headers.get(\"ratelimit\"));\n\t\t\t\tthrow new Error(\n\t\t\t\t\tdelay ? `Hugging Face rate limit reached; retry in ${delay}s` : \"Hugging Face rate limit reached\",\n\t\t\t\t);\n\t\t\t}\n\t\t\tthrow new Error(payloadError(payload, fallback));\n\t\t}\n\t\treturn payload;\n\t}\n\n\tasync search(query: string, signal?: AbortSignal): Promise<HuggingFaceModel[]> {\n\t\tconst params = new URLSearchParams({\n\t\t\tsearch: query,\n\t\t\tfilter: \"gguf\",\n\t\t\tsort: \"downloads\",\n\t\t\tdirection: \"-1\",\n\t\t\tlimit: \"20\",\n\t\t});\n\t\tconst payload = await this.request(`/api/models?${params}`, signal);\n\t\tif (!Array.isArray(payload)) throw new Error(\"Hugging Face returned invalid search results\");\n\t\treturn payload.flatMap((value) => {\n\t\t\tif (typeof value !== \"object\" || value === null || typeof (value as { id?: unknown }).id !== \"string\")\n\t\t\t\treturn [];\n\t\t\tconst model = value as { id: string; downloads?: unknown };\n\t\t\treturn [{ id: model.id, downloads: typeof model.downloads === \"number\" ? model.downloads : 0 }];\n\t\t});\n\t}\n\n\tasync details(id: string, signal?: AbortSignal): Promise<HuggingFaceModelDetails> {\n\t\tconst encodedId = id.split(\"/\").map(encodeURIComponent).join(\"/\");\n\t\tconst payload = await this.request(`/api/models/${encodedId}?blobs=true`, signal);\n\t\tif (typeof payload !== \"object\" || payload === null) {\n\t\t\tthrow new Error(\"Hugging Face returned invalid model details\");\n\t\t}\n\t\tconst model = payload as { id?: unknown; gated?: unknown; siblings?: unknown };\n\t\tconst sizes = new Map<string, { total: number; complete: boolean }>();\n\t\tif (Array.isArray(model.siblings)) {\n\t\t\tfor (const value of model.siblings) {\n\t\t\t\tif (typeof value !== \"object\" || value === null) continue;\n\t\t\t\tconst file = value as { rfilename?: unknown; size?: unknown };\n\t\t\t\tif (typeof file.rfilename !== \"string\" || !file.rfilename.toLowerCase().endsWith(\".gguf\")) continue;\n\t\t\t\tconst filename = file.rfilename.split(\"/\").at(-1)!;\n\t\t\t\tif (filename.toLowerCase().startsWith(\"mmproj\")) continue;\n\t\t\t\tconst stem = filename.slice(0, -5).replace(SHARD_SUFFIX_PATTERN, \"\");\n\t\t\t\tconst quantization = stem.match(QUANTIZATION_PATTERN)?.[1]?.toUpperCase();\n\t\t\t\tif (!quantization) continue;\n\t\t\t\tconst current = sizes.get(quantization) ?? { total: 0, complete: true };\n\t\t\t\tif (typeof file.size === \"number\") current.total += file.size;\n\t\t\t\telse current.complete = false;\n\t\t\t\tsizes.set(quantization, current);\n\t\t\t}\n\t\t}\n\t\tconst quantizations = [...sizes]\n\t\t\t.map(([name, size]) => ({ name, size: size.complete ? size.total : undefined }))\n\t\t\t.sort((left, right) => {\n\t\t\t\tif (left.name === \"Q4_K_M\") return -1;\n\t\t\t\tif (right.name === \"Q4_K_M\") return 1;\n\t\t\t\treturn (\n\t\t\t\t\t(left.size ?? Number.MAX_SAFE_INTEGER) - (right.size ?? Number.MAX_SAFE_INTEGER) ||\n\t\t\t\t\tleft.name.localeCompare(right.name)\n\t\t\t\t);\n\t\t\t});\n\t\treturn {\n\t\t\tid: typeof model.id === \"string\" ? model.id : id,\n\t\t\tgated: model.gated === \"auto\" || model.gated === \"manual\" ? model.gated : false,\n\t\t\tquantizations,\n\t\t};\n\t}\n}\n"]}
@@ -1,3 +0,0 @@
1
- import type { ExtensionAPI } from "../../core/extensions/types.ts";
2
- export default function llamaExtension(pi: ExtensionAPI): void;
3
- //# sourceMappingURL=index.d.ts.map
@@ -1 +0,0 @@
1
- {"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../../src/extensions/llama/index.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,YAAY,EAA2B,MAAM,gCAAgC,CAAC;AAyC5F,MAAM,CAAC,OAAO,UAAU,cAAc,CAAC,EAAE,EAAE,YAAY,GAAG,IAAI,CAmL7D","sourcesContent":["import type { ExtensionAPI, ExtensionCommandContext } from \"../../core/extensions/types.ts\";\nimport { formatBytes, LlamaClient, type LlamaModelInfo, normalizeLlamaServerUrl } from \"./client.ts\";\nimport { findHuggingFaceToken, HuggingFaceClient } from \"./huggingface.ts\";\nimport { createLlamaProvider, LLAMA_PROVIDER_ID } from \"./provider.ts\";\nimport { type LlamaUi, runWithProgress, showLlamaUi } from \"./ui.ts\";\n\nfunction modelIsLoaded(model: LlamaModelInfo): boolean {\n\treturn model.status.value === \"loaded\" || model.status.value === \"sleeping\";\n}\n\nfunction isConnectionError(error: unknown): boolean {\n\tif (!(error instanceof Error)) return false;\n\tconst message = `${error.name} ${error.message}`.toLowerCase();\n\treturn message.includes(\"fetch failed\") || message.includes(\"timeout\") || message.includes(\"network\");\n}\n\nfunction connectionErrorMessage(error: unknown): string {\n\tif (isConnectionError(error)) return \"Could not connect to the server.\";\n\treturn error instanceof Error ? error.message : String(error);\n}\n\nfunction parseHuggingFaceModel(value: string): { repository: string; quantization?: string } {\n\tconst colon = value.indexOf(\":\", value.indexOf(\"/\") + 1);\n\treturn colon < 0\n\t\t? { repository: value }\n\t\t: { repository: value.slice(0, colon), quantization: value.slice(colon + 1) };\n}\n\nasync function configuredClient(ctx: ExtensionCommandContext): Promise<LlamaClient | undefined> {\n\tconst result = await ctx.modelRegistry.getProviderAuth(LLAMA_PROVIDER_ID);\n\tif (!result) {\n\t\tctx.ui.notify(`Configure llama.cpp with /login ${LLAMA_PROVIDER_ID}`, \"warning\");\n\t\treturn undefined;\n\t}\n\tconst configuredUrl = result.env?.LLAMA_BASE_URL;\n\tconst serverUrl = normalizeLlamaServerUrl(\n\t\ttypeof configuredUrl === \"string\" && configuredUrl ? configuredUrl : (result.auth.baseUrl ?? \"\"),\n\t);\n\treturn new LlamaClient(serverUrl, result.auth.apiKey);\n}\n\nexport default function llamaExtension(pi: ExtensionAPI): void {\n\tconst provider = createLlamaProvider();\n\tpi.registerProvider(provider.provider);\n\n\tconst syncCatalog = async (\n\t\tctx: ExtensionCommandContext,\n\t\tclient: LlamaClient,\n\t\tcatalog?: LlamaModelInfo[],\n\t): Promise<LlamaModelInfo[]> => {\n\t\tconst current = catalog ?? (await client.list());\n\t\tprovider.setCatalog(current, client.serverUrl);\n\t\tawait ctx.modelRegistry.refresh();\n\t\treturn current;\n\t};\n\n\tconst loadModel = async (\n\t\tctx: ExtensionCommandContext,\n\t\tui: LlamaUi,\n\t\tclient: LlamaClient,\n\t\tcatalog: LlamaModelInfo[],\n\t\ttarget: LlamaModelInfo,\n\t): Promise<void> => {\n\t\tconst loaded = catalog.filter((model) => model.id !== target.id && modelIsLoaded(model));\n\t\tlet replace = false;\n\t\tif (loaded.length > 0) {\n\t\t\tconst choice = await ui.select(`${loaded.length} model${loaded.length === 1 ? \" is\" : \"s are\"} loaded`, [\n\t\t\t\t\"Unload all and load\",\n\t\t\t\t\"Keep loaded and load\",\n\t\t\t\t\"Cancel\",\n\t\t\t]);\n\t\t\tif (!choice || choice === \"Cancel\") return;\n\t\t\treplace = choice === \"Unload all and load\";\n\t\t}\n\n\t\tconst restoreLoaded = async (): Promise<void> => {\n\t\t\tctx.ui.notify(\"Restoring previously loaded models\");\n\t\t\tfor (const model of loaded) await client.loadAndWait(model.id, () => {});\n\t\t\tawait syncCatalog(ctx, client);\n\t\t};\n\t\tif (replace) {\n\t\t\tfor (const model of loaded) await client.unloadAndWait(model.id);\n\t\t}\n\n\t\ttry {\n\t\t\tconst result = await runWithProgress(ui, {\n\t\t\t\ttitle: \"Loading model\",\n\t\t\t\tmodel: target.id,\n\t\t\t\tinitialMessage: \"Starting…\",\n\t\t\t\tcancelTitle: \"Stop loading?\",\n\t\t\t\tcancelMessage: target.id,\n\t\t\t\trun: (signal, update) => client.loadAndWait(target.id, update, signal),\n\t\t\t\tcancel: () => client.unload(target.id),\n\t\t\t});\n\t\t\tif (result.cancelled) {\n\t\t\t\tif (replace) await restoreLoaded();\n\t\t\t\treturn;\n\t\t\t}\n\t\t\tconst refreshed = await syncCatalog(ctx, client);\n\t\t\tconst loadedModel = refreshed.find((model) => model.id === target.id);\n\t\t\tctx.ui.notify(\n\t\t\t\tloadedModel?.status.value === \"loaded\" ? `Loaded ${target.id}` : `Load started for ${target.id}`,\n\t\t\t);\n\t\t} catch (error) {\n\t\t\tif (replace) {\n\t\t\t\ttry {\n\t\t\t\t\tawait restoreLoaded();\n\t\t\t\t} catch {\n\t\t\t\t\t// Preserve the original load error.\n\t\t\t\t}\n\t\t\t}\n\t\t\tthrow error;\n\t\t}\n\t};\n\n\tconst unloadModel = async (\n\t\tctx: ExtensionCommandContext,\n\t\tui: LlamaUi,\n\t\tclient: LlamaClient,\n\t\tmodel: LlamaModelInfo,\n\t): Promise<void> => {\n\t\tif (!(await ui.confirm(\"Unload model?\", model.id))) return;\n\t\tawait client.unloadAndWait(model.id);\n\t\tawait syncCatalog(ctx, client);\n\t\tctx.ui.notify(`Unloaded ${model.id}`);\n\t};\n\n\tconst downloadModel = async (ctx: ExtensionCommandContext, ui: LlamaUi, client: LlamaClient): Promise<void> => {\n\t\tconst huggingFace = new HuggingFaceClient(await findHuggingFaceToken());\n\t\tconst selected = await ui.searchModels((query, signal) => huggingFace.search(query, signal));\n\t\tif (!selected) return;\n\t\tconst parsed = parseHuggingFaceModel(selected);\n\t\tui.showStatus(\"Loading model details\", parsed.repository);\n\t\tconst details = await huggingFace.details(parsed.repository);\n\t\tif (details.gated) {\n\t\t\tconst approval = details.gated === \"manual\" ? \"Manual approval is required\" : \"Accept the access terms\";\n\t\t\tconst choice = await ui.select(\n\t\t\t\t`Hugging Face access required\\n${details.id}\\n\\n${approval} at:\\nhttps://huggingface.co/${details.id}\\n\\nThe llama.cpp server needs HF_TOKEN with access.`,\n\t\t\t\t[\"Continue\", \"Back\"],\n\t\t\t);\n\t\t\tif (choice !== \"Continue\") return;\n\t\t}\n\t\tlet quantization = parsed.quantization;\n\t\tif (!quantization && details.quantizations.length > 0) {\n\t\t\tconst options = details.quantizations.map((entry) => {\n\t\t\t\tconst detail = [\n\t\t\t\t\tentry.size === undefined ? undefined : formatBytes(entry.size),\n\t\t\t\t\tentry.name === \"Q4_K_M\" ? \"recommended\" : undefined,\n\t\t\t\t]\n\t\t\t\t\t.filter((value): value is string => Boolean(value))\n\t\t\t\t\t.join(\" · \");\n\t\t\t\treturn detail ? `${entry.name} · ${detail}` : entry.name;\n\t\t\t});\n\t\t\tconst choice = await ui.select(`Select quantization\\n${details.id}`, options);\n\t\t\tif (!choice) return;\n\t\t\tquantization = details.quantizations[options.indexOf(choice)]?.name;\n\t\t\tif (!quantization) return;\n\t\t}\n\t\tconst model = quantization ? `${details.id}:${quantization}` : details.id;\n\t\tconst result = await runWithProgress(ui, {\n\t\t\ttitle: \"Downloading model\",\n\t\t\tmodel,\n\t\t\tinitialMessage: \"Starting…\",\n\t\t\tcancelTitle: \"Stop download?\",\n\t\t\tcancelMessage: model,\n\t\t\trun: (signal, update) => client.downloadAndWait(model, update, signal),\n\t\t\tcancel: () => client.unload(model),\n\t\t});\n\t\tif (result.cancelled) return;\n\t\tawait syncCatalog(ctx, client, result.value);\n\t\tctx.ui.notify(`Downloaded ${model}`);\n\t};\n\n\tpi.registerCommand(\"llama\", {\n\t\tdescription: \"Manage llama.cpp router models\",\n\t\thandler: async (_args, ctx) => {\n\t\t\tif (ctx.mode !== \"tui\") {\n\t\t\t\tctx.ui.notify(\"/llama is available in interactive mode\", \"warning\");\n\t\t\t\treturn;\n\t\t\t}\n\t\t\tconst client = await configuredClient(ctx);\n\t\t\tif (!client) return;\n\t\t\tawait showLlamaUi(ctx, async (ui) => {\n\t\t\t\tconst readCatalog = async (): Promise<LlamaModelInfo[] | undefined> => {\n\t\t\t\t\twhile (true) {\n\t\t\t\t\t\ttry {\n\t\t\t\t\t\t\treturn await syncCatalog(ctx, client);\n\t\t\t\t\t\t} catch (error) {\n\t\t\t\t\t\t\tif ((await ui.connectionError(client.serverUrl, connectionErrorMessage(error))) === \"close\") {\n\t\t\t\t\t\t\t\treturn undefined;\n\t\t\t\t\t\t\t}\n\t\t\t\t\t\t}\n\t\t\t\t\t}\n\t\t\t\t};\n\n\t\t\t\tlet catalog = await readCatalog();\n\t\t\t\tif (!catalog) return;\n\t\t\t\twhile (true) {\n\t\t\t\t\tconst action = await ui.showModels(client.serverUrl, catalog);\n\t\t\t\t\tif (action.type === \"close\") return;\n\t\t\t\t\tlet actionError: unknown;\n\t\t\t\t\ttry {\n\t\t\t\t\t\tif (action.type === \"download\") await downloadModel(ctx, ui, client);\n\t\t\t\t\t\telse if (modelIsLoaded(action.model)) await unloadModel(ctx, ui, client, action.model);\n\t\t\t\t\t\telse if (action.model.status.value === \"unloaded\")\n\t\t\t\t\t\t\tawait loadModel(ctx, ui, client, catalog, action.model);\n\t\t\t\t\t\telse ctx.ui.notify(`${action.model.id} is ${action.model.status.value}`, \"warning\");\n\t\t\t\t\t} catch (error) {\n\t\t\t\t\t\tactionError = error;\n\t\t\t\t\t}\n\t\t\t\t\tconst refreshed = await readCatalog();\n\t\t\t\t\tif (!refreshed) return;\n\t\t\t\t\tcatalog = refreshed;\n\t\t\t\t\tif (actionError && !isConnectionError(actionError)) {\n\t\t\t\t\t\tctx.ui.notify(actionError instanceof Error ? actionError.message : String(actionError), \"error\");\n\t\t\t\t\t}\n\t\t\t\t}\n\t\t\t});\n\t\t},\n\t});\n}\n"]}
@@ -1,211 +0,0 @@
1
- "use strict";
2
- Object.defineProperty(exports, "__esModule", { value: true });
3
- exports.default = llamaExtension;
4
- const client_ts_1 = require("./client.js");
5
- const huggingface_ts_1 = require("./huggingface.js");
6
- const provider_ts_1 = require("./provider.js");
7
- const ui_ts_1 = require("./ui.js");
8
- function modelIsLoaded(model) {
9
- return model.status.value === "loaded" || model.status.value === "sleeping";
10
- }
11
- function isConnectionError(error) {
12
- if (!(error instanceof Error))
13
- return false;
14
- const message = `${error.name} ${error.message}`.toLowerCase();
15
- return message.includes("fetch failed") || message.includes("timeout") || message.includes("network");
16
- }
17
- function connectionErrorMessage(error) {
18
- if (isConnectionError(error))
19
- return "Could not connect to the server.";
20
- return error instanceof Error ? error.message : String(error);
21
- }
22
- function parseHuggingFaceModel(value) {
23
- const colon = value.indexOf(":", value.indexOf("/") + 1);
24
- return colon < 0
25
- ? { repository: value }
26
- : { repository: value.slice(0, colon), quantization: value.slice(colon + 1) };
27
- }
28
- async function configuredClient(ctx) {
29
- const result = await ctx.modelRegistry.getProviderAuth(provider_ts_1.LLAMA_PROVIDER_ID);
30
- if (!result) {
31
- ctx.ui.notify(`Configure llama.cpp with /login ${provider_ts_1.LLAMA_PROVIDER_ID}`, "warning");
32
- return undefined;
33
- }
34
- const configuredUrl = result.env?.LLAMA_BASE_URL;
35
- const serverUrl = (0, client_ts_1.normalizeLlamaServerUrl)(typeof configuredUrl === "string" && configuredUrl ? configuredUrl : (result.auth.baseUrl ?? ""));
36
- return new client_ts_1.LlamaClient(serverUrl, result.auth.apiKey);
37
- }
38
- function llamaExtension(pi) {
39
- const provider = (0, provider_ts_1.createLlamaProvider)();
40
- pi.registerProvider(provider.provider);
41
- const syncCatalog = async (ctx, client, catalog) => {
42
- const current = catalog ?? (await client.list());
43
- provider.setCatalog(current, client.serverUrl);
44
- await ctx.modelRegistry.refresh();
45
- return current;
46
- };
47
- const loadModel = async (ctx, ui, client, catalog, target) => {
48
- const loaded = catalog.filter((model) => model.id !== target.id && modelIsLoaded(model));
49
- let replace = false;
50
- if (loaded.length > 0) {
51
- const choice = await ui.select(`${loaded.length} model${loaded.length === 1 ? " is" : "s are"} loaded`, [
52
- "Unload all and load",
53
- "Keep loaded and load",
54
- "Cancel",
55
- ]);
56
- if (!choice || choice === "Cancel")
57
- return;
58
- replace = choice === "Unload all and load";
59
- }
60
- const restoreLoaded = async () => {
61
- ctx.ui.notify("Restoring previously loaded models");
62
- for (const model of loaded)
63
- await client.loadAndWait(model.id, () => { });
64
- await syncCatalog(ctx, client);
65
- };
66
- if (replace) {
67
- for (const model of loaded)
68
- await client.unloadAndWait(model.id);
69
- }
70
- try {
71
- const result = await (0, ui_ts_1.runWithProgress)(ui, {
72
- title: "Loading model",
73
- model: target.id,
74
- initialMessage: "Starting…",
75
- cancelTitle: "Stop loading?",
76
- cancelMessage: target.id,
77
- run: (signal, update) => client.loadAndWait(target.id, update, signal),
78
- cancel: () => client.unload(target.id),
79
- });
80
- if (result.cancelled) {
81
- if (replace)
82
- await restoreLoaded();
83
- return;
84
- }
85
- const refreshed = await syncCatalog(ctx, client);
86
- const loadedModel = refreshed.find((model) => model.id === target.id);
87
- ctx.ui.notify(loadedModel?.status.value === "loaded" ? `Loaded ${target.id}` : `Load started for ${target.id}`);
88
- }
89
- catch (error) {
90
- if (replace) {
91
- try {
92
- await restoreLoaded();
93
- }
94
- catch {
95
- // Preserve the original load error.
96
- }
97
- }
98
- throw error;
99
- }
100
- };
101
- const unloadModel = async (ctx, ui, client, model) => {
102
- if (!(await ui.confirm("Unload model?", model.id)))
103
- return;
104
- await client.unloadAndWait(model.id);
105
- await syncCatalog(ctx, client);
106
- ctx.ui.notify(`Unloaded ${model.id}`);
107
- };
108
- const downloadModel = async (ctx, ui, client) => {
109
- const huggingFace = new huggingface_ts_1.HuggingFaceClient(await (0, huggingface_ts_1.findHuggingFaceToken)());
110
- const selected = await ui.searchModels((query, signal) => huggingFace.search(query, signal));
111
- if (!selected)
112
- return;
113
- const parsed = parseHuggingFaceModel(selected);
114
- ui.showStatus("Loading model details", parsed.repository);
115
- const details = await huggingFace.details(parsed.repository);
116
- if (details.gated) {
117
- const approval = details.gated === "manual" ? "Manual approval is required" : "Accept the access terms";
118
- const choice = await ui.select(`Hugging Face access required\n${details.id}\n\n${approval} at:\nhttps://huggingface.co/${details.id}\n\nThe llama.cpp server needs HF_TOKEN with access.`, ["Continue", "Back"]);
119
- if (choice !== "Continue")
120
- return;
121
- }
122
- let quantization = parsed.quantization;
123
- if (!quantization && details.quantizations.length > 0) {
124
- const options = details.quantizations.map((entry) => {
125
- const detail = [
126
- entry.size === undefined ? undefined : (0, client_ts_1.formatBytes)(entry.size),
127
- entry.name === "Q4_K_M" ? "recommended" : undefined,
128
- ]
129
- .filter((value) => Boolean(value))
130
- .join(" · ");
131
- return detail ? `${entry.name} · ${detail}` : entry.name;
132
- });
133
- const choice = await ui.select(`Select quantization\n${details.id}`, options);
134
- if (!choice)
135
- return;
136
- quantization = details.quantizations[options.indexOf(choice)]?.name;
137
- if (!quantization)
138
- return;
139
- }
140
- const model = quantization ? `${details.id}:${quantization}` : details.id;
141
- const result = await (0, ui_ts_1.runWithProgress)(ui, {
142
- title: "Downloading model",
143
- model,
144
- initialMessage: "Starting…",
145
- cancelTitle: "Stop download?",
146
- cancelMessage: model,
147
- run: (signal, update) => client.downloadAndWait(model, update, signal),
148
- cancel: () => client.unload(model),
149
- });
150
- if (result.cancelled)
151
- return;
152
- await syncCatalog(ctx, client, result.value);
153
- ctx.ui.notify(`Downloaded ${model}`);
154
- };
155
- pi.registerCommand("llama", {
156
- description: "Manage llama.cpp router models",
157
- handler: async (_args, ctx) => {
158
- if (ctx.mode !== "tui") {
159
- ctx.ui.notify("/llama is available in interactive mode", "warning");
160
- return;
161
- }
162
- const client = await configuredClient(ctx);
163
- if (!client)
164
- return;
165
- await (0, ui_ts_1.showLlamaUi)(ctx, async (ui) => {
166
- const readCatalog = async () => {
167
- while (true) {
168
- try {
169
- return await syncCatalog(ctx, client);
170
- }
171
- catch (error) {
172
- if ((await ui.connectionError(client.serverUrl, connectionErrorMessage(error))) === "close") {
173
- return undefined;
174
- }
175
- }
176
- }
177
- };
178
- let catalog = await readCatalog();
179
- if (!catalog)
180
- return;
181
- while (true) {
182
- const action = await ui.showModels(client.serverUrl, catalog);
183
- if (action.type === "close")
184
- return;
185
- let actionError;
186
- try {
187
- if (action.type === "download")
188
- await downloadModel(ctx, ui, client);
189
- else if (modelIsLoaded(action.model))
190
- await unloadModel(ctx, ui, client, action.model);
191
- else if (action.model.status.value === "unloaded")
192
- await loadModel(ctx, ui, client, catalog, action.model);
193
- else
194
- ctx.ui.notify(`${action.model.id} is ${action.model.status.value}`, "warning");
195
- }
196
- catch (error) {
197
- actionError = error;
198
- }
199
- const refreshed = await readCatalog();
200
- if (!refreshed)
201
- return;
202
- catalog = refreshed;
203
- if (actionError && !isConnectionError(actionError)) {
204
- ctx.ui.notify(actionError instanceof Error ? actionError.message : String(actionError), "error");
205
- }
206
- }
207
- });
208
- },
209
- });
210
- }
211
- //# sourceMappingURL=index.js.map
@@ -1 +0,0 @@
1
- {"version":3,"file":"index.js","sourceRoot":"","sources":["../../../src/extensions/llama/index.ts"],"names":[],"mappings":";;;AACA,2CAAqG;AACrG,qDAA2E;AAC3E,+CAAuE;AACvE,mCAAqE;AAErE,SAAS,aAAa,CAAC,KAAqB,EAAW;IACtD,OAAO,KAAK,CAAC,MAAM,CAAC,KAAK,KAAK,QAAQ,IAAI,KAAK,CAAC,MAAM,CAAC,KAAK,KAAK,UAAU,CAAC;AAAA,CAC5E;AAED,SAAS,iBAAiB,CAAC,KAAc,EAAW;IACnD,IAAI,CAAC,CAAC,KAAK,YAAY,KAAK,CAAC;QAAE,OAAO,KAAK,CAAC;IAC5C,MAAM,OAAO,GAAG,GAAG,KAAK,CAAC,IAAI,IAAI,KAAK,CAAC,OAAO,EAAE,CAAC,WAAW,EAAE,CAAC;IAC/D,OAAO,OAAO,CAAC,QAAQ,CAAC,cAAc,CAAC,IAAI,OAAO,CAAC,QAAQ,CAAC,SAAS,CAAC,IAAI,OAAO,CAAC,QAAQ,CAAC,SAAS,CAAC,CAAC;AAAA,CACtG;AAED,SAAS,sBAAsB,CAAC,KAAc,EAAU;IACvD,IAAI,iBAAiB,CAAC,KAAK,CAAC;QAAE,OAAO,kCAAkC,CAAC;IACxE,OAAO,KAAK,YAAY,KAAK,CAAC,CAAC,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC;AAAA,CAC9D;AAED,SAAS,qBAAqB,CAAC,KAAa,EAAiD;IAC5F,MAAM,KAAK,GAAG,KAAK,CAAC,OAAO,CAAC,GAAG,EAAE,KAAK,CAAC,OAAO,CAAC,GAAG,CAAC,GAAG,CAAC,CAAC,CAAC;IACzD,OAAO,KAAK,GAAG,CAAC;QACf,CAAC,CAAC,EAAE,UAAU,EAAE,KAAK,EAAE;QACvB,CAAC,CAAC,EAAE,UAAU,EAAE,KAAK,CAAC,KAAK,CAAC,CAAC,EAAE,KAAK,CAAC,EAAE,YAAY,EAAE,KAAK,CAAC,KAAK,CAAC,KAAK,GAAG,CAAC,CAAC,EAAE,CAAC;AAAA,CAC/E;AAED,KAAK,UAAU,gBAAgB,CAAC,GAA4B,EAAoC;IAC/F,MAAM,MAAM,GAAG,MAAM,GAAG,CAAC,aAAa,CAAC,eAAe,CAAC,+BAAiB,CAAC,CAAC;IAC1E,IAAI,CAAC,MAAM,EAAE,CAAC;QACb,GAAG,CAAC,EAAE,CAAC,MAAM,CAAC,mCAAmC,+BAAiB,EAAE,EAAE,SAAS,CAAC,CAAC;QACjF,OAAO,SAAS,CAAC;IAClB,CAAC;IACD,MAAM,aAAa,GAAG,MAAM,CAAC,GAAG,EAAE,cAAc,CAAC;IACjD,MAAM,SAAS,GAAG,IAAA,mCAAuB,EACxC,OAAO,aAAa,KAAK,QAAQ,IAAI,aAAa,CAAC,CAAC,CAAC,aAAa,CAAC,CAAC,CAAC,CAAC,MAAM,CAAC,IAAI,CAAC,OAAO,IAAI,EAAE,CAAC,CAChG,CAAC;IACF,OAAO,IAAI,uBAAW,CAAC,SAAS,EAAE,MAAM,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC;AAAA,CACtD;AAED,wBAAuC,EAAgB,EAAQ;IAC9D,MAAM,QAAQ,GAAG,IAAA,iCAAmB,GAAE,CAAC;IACvC,EAAE,CAAC,gBAAgB,CAAC,QAAQ,CAAC,QAAQ,CAAC,CAAC;IAEvC,MAAM,WAAW,GAAG,KAAK,EACxB,GAA4B,EAC5B,MAAmB,EACnB,OAA0B,EACE,EAAE,CAAC;QAC/B,MAAM,OAAO,GAAG,OAAO,IAAI,CAAC,MAAM,MAAM,CAAC,IAAI,EAAE,CAAC,CAAC;QACjD,QAAQ,CAAC,UAAU,CAAC,OAAO,EAAE,MAAM,CAAC,SAAS,CAAC,CAAC;QAC/C,MAAM,GAAG,CAAC,aAAa,CAAC,OAAO,EAAE,CAAC;QAClC,OAAO,OAAO,CAAC;IAAA,CACf,CAAC;IAEF,MAAM,SAAS,GAAG,KAAK,EACtB,GAA4B,EAC5B,EAAW,EACX,MAAmB,EACnB,OAAyB,EACzB,MAAsB,EACN,EAAE,CAAC;QACnB,MAAM,MAAM,GAAG,OAAO,CAAC,MAAM,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,KAAK,CAAC,EAAE,KAAK,MAAM,CAAC,EAAE,IAAI,aAAa,CAAC,KAAK,CAAC,CAAC,CAAC;QACzF,IAAI,OAAO,GAAG,KAAK,CAAC;QACpB,IAAI,MAAM,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;YACvB,MAAM,MAAM,GAAG,MAAM,EAAE,CAAC,MAAM,CAAC,GAAG,MAAM,CAAC,MAAM,SAAS,MAAM,CAAC,MAAM,KAAK,CAAC,CAAC,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,OAAO,SAAS,EAAE;gBACvG,qBAAqB;gBACrB,sBAAsB;gBACtB,QAAQ;aACR,CAAC,CAAC;YACH,IAAI,CAAC,MAAM,IAAI,MAAM,KAAK,QAAQ;gBAAE,OAAO;YAC3C,OAAO,GAAG,MAAM,KAAK,qBAAqB,CAAC;QAC5C,CAAC;QAED,MAAM,aAAa,GAAG,KAAK,IAAmB,EAAE,CAAC;YAChD,GAAG,CAAC,EAAE,CAAC,MAAM,CAAC,oCAAoC,CAAC,CAAC;YACpD,KAAK,MAAM,KAAK,IAAI,MAAM;gBAAE,MAAM,MAAM,CAAC,WAAW,CAAC,KAAK,CAAC,EAAE,EAAE,GAAG,EAAE,CAAC,EAAC,CAAC,CAAC,CAAC;YACzE,MAAM,WAAW,CAAC,GAAG,EAAE,MAAM,CAAC,CAAC;QAAA,CAC/B,CAAC;QACF,IAAI,OAAO,EAAE,CAAC;YACb,KAAK,MAAM,KAAK,IAAI,MAAM;gBAAE,MAAM,MAAM,CAAC,aAAa,CAAC,KAAK,CAAC,EAAE,CAAC,CAAC;QAClE,CAAC;QAED,IAAI,CAAC;YACJ,MAAM,MAAM,GAAG,MAAM,IAAA,uBAAe,EAAC,EAAE,EAAE;gBACxC,KAAK,EAAE,eAAe;gBACtB,KAAK,EAAE,MAAM,CAAC,EAAE;gBAChB,cAAc,EAAE,aAAW;gBAC3B,WAAW,EAAE,eAAe;gBAC5B,aAAa,EAAE,MAAM,CAAC,EAAE;gBACxB,GAAG,EAAE,CAAC,MAAM,EAAE,MAAM,EAAE,EAAE,CAAC,MAAM,CAAC,WAAW,CAAC,MAAM,CAAC,EAAE,EAAE,MAAM,EAAE,MAAM,CAAC;gBACtE,MAAM,EAAE,GAAG,EAAE,CAAC,MAAM,CAAC,MAAM,CAAC,MAAM,CAAC,EAAE,CAAC;aACtC,CAAC,CAAC;YACH,IAAI,MAAM,CAAC,SAAS,EAAE,CAAC;gBACtB,IAAI,OAAO;oBAAE,MAAM,aAAa,EAAE,CAAC;gBACnC,OAAO;YACR,CAAC;YACD,MAAM,SAAS,GAAG,MAAM,WAAW,CAAC,GAAG,EAAE,MAAM,CAAC,CAAC;YACjD,MAAM,WAAW,GAAG,SAAS,CAAC,IAAI,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,KAAK,CAAC,EAAE,KAAK,MAAM,CAAC,EAAE,CAAC,CAAC;YACtE,GAAG,CAAC,EAAE,CAAC,MAAM,CACZ,WAAW,EAAE,MAAM,CAAC,KAAK,KAAK,QAAQ,CAAC,CAAC,CAAC,UAAU,MAAM,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,oBAAoB,MAAM,CAAC,EAAE,EAAE,CAChG,CAAC;QACH,CAAC;QAAC,OAAO,KAAK,EAAE,CAAC;YAChB,IAAI,OAAO,EAAE,CAAC;gBACb,IAAI,CAAC;oBACJ,MAAM,aAAa,EAAE,CAAC;gBACvB,CAAC;gBAAC,MAAM,CAAC;oBACR,oCAAoC;gBACrC,CAAC;YACF,CAAC;YACD,MAAM,KAAK,CAAC;QACb,CAAC;IAAA,CACD,CAAC;IAEF,MAAM,WAAW,GAAG,KAAK,EACxB,GAA4B,EAC5B,EAAW,EACX,MAAmB,EACnB,KAAqB,EACL,EAAE,CAAC;QACnB,IAAI,CAAC,CAAC,MAAM,EAAE,CAAC,OAAO,CAAC,eAAe,EAAE,KAAK,CAAC,EAAE,CAAC,CAAC;YAAE,OAAO;QAC3D,MAAM,MAAM,CAAC,aAAa,CAAC,KAAK,CAAC,EAAE,CAAC,CAAC;QACrC,MAAM,WAAW,CAAC,GAAG,EAAE,MAAM,CAAC,CAAC;QAC/B,GAAG,CAAC,EAAE,CAAC,MAAM,CAAC,YAAY,KAAK,CAAC,EAAE,EAAE,CAAC,CAAC;IAAA,CACtC,CAAC;IAEF,MAAM,aAAa,GAAG,KAAK,EAAE,GAA4B,EAAE,EAAW,EAAE,MAAmB,EAAiB,EAAE,CAAC;QAC9G,MAAM,WAAW,GAAG,IAAI,kCAAiB,CAAC,MAAM,IAAA,qCAAoB,GAAE,CAAC,CAAC;QACxE,MAAM,QAAQ,GAAG,MAAM,EAAE,CAAC,YAAY,CAAC,CAAC,KAAK,EAAE,MAAM,EAAE,EAAE,CAAC,WAAW,CAAC,MAAM,CAAC,KAAK,EAAE,MAAM,CAAC,CAAC,CAAC;QAC7F,IAAI,CAAC,QAAQ;YAAE,OAAO;QACtB,MAAM,MAAM,GAAG,qBAAqB,CAAC,QAAQ,CAAC,CAAC;QAC/C,EAAE,CAAC,UAAU,CAAC,uBAAuB,EAAE,MAAM,CAAC,UAAU,CAAC,CAAC;QAC1D,MAAM,OAAO,GAAG,MAAM,WAAW,CAAC,OAAO,CAAC,MAAM,CAAC,UAAU,CAAC,CAAC;QAC7D,IAAI,OAAO,CAAC,KAAK,EAAE,CAAC;YACnB,MAAM,QAAQ,GAAG,OAAO,CAAC,KAAK,KAAK,QAAQ,CAAC,CAAC,CAAC,6BAA6B,CAAC,CAAC,CAAC,yBAAyB,CAAC;YACxG,MAAM,MAAM,GAAG,MAAM,EAAE,CAAC,MAAM,CAC7B,iCAAiC,OAAO,CAAC,EAAE,OAAO,QAAQ,gCAAgC,OAAO,CAAC,EAAE,sDAAsD,EAC1J,CAAC,UAAU,EAAE,MAAM,CAAC,CACpB,CAAC;YACF,IAAI,MAAM,KAAK,UAAU;gBAAE,OAAO;QACnC,CAAC;QACD,IAAI,YAAY,GAAG,MAAM,CAAC,YAAY,CAAC;QACvC,IAAI,CAAC,YAAY,IAAI,OAAO,CAAC,aAAa,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;YACvD,MAAM,OAAO,GAAG,OAAO,CAAC,aAAa,CAAC,GAAG,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC;gBACpD,MAAM,MAAM,GAAG;oBACd,KAAK,CAAC,IAAI,KAAK,SAAS,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,IAAA,uBAAW,EAAC,KAAK,CAAC,IAAI,CAAC;oBAC9D,KAAK,CAAC,IAAI,KAAK,QAAQ,CAAC,CAAC,CAAC,aAAa,CAAC,CAAC,CAAC,SAAS;iBACnD;qBACC,MAAM,CAAC,CAAC,KAAK,EAAmB,EAAE,CAAC,OAAO,CAAC,KAAK,CAAC,CAAC;qBAClD,IAAI,CAAC,MAAK,CAAC,CAAC;gBACd,OAAO,MAAM,CAAC,CAAC,CAAC,GAAG,KAAK,CAAC,IAAI,OAAM,MAAM,EAAE,CAAC,CAAC,CAAC,KAAK,CAAC,IAAI,CAAC;YAAA,CACzD,CAAC,CAAC;YACH,MAAM,MAAM,GAAG,MAAM,EAAE,CAAC,MAAM,CAAC,wBAAwB,OAAO,CAAC,EAAE,EAAE,EAAE,OAAO,CAAC,CAAC;YAC9E,IAAI,CAAC,MAAM;gBAAE,OAAO;YACpB,YAAY,GAAG,OAAO,CAAC,aAAa,CAAC,OAAO,CAAC,OAAO,CAAC,MAAM,CAAC,CAAC,EAAE,IAAI,CAAC;YACpE,IAAI,CAAC,YAAY;gBAAE,OAAO;QAC3B,CAAC;QACD,MAAM,KAAK,GAAG,YAAY,CAAC,CAAC,CAAC,GAAG,OAAO,CAAC,EAAE,IAAI,YAAY,EAAE,CAAC,CAAC,CAAC,OAAO,CAAC,EAAE,CAAC;QAC1E,MAAM,MAAM,GAAG,MAAM,IAAA,uBAAe,EAAC,EAAE,EAAE;YACxC,KAAK,EAAE,mBAAmB;YAC1B,KAAK;YACL,cAAc,EAAE,aAAW;YAC3B,WAAW,EAAE,gBAAgB;YAC7B,aAAa,EAAE,KAAK;YACpB,GAAG,EAAE,CAAC,MAAM,EAAE,MAAM,EAAE,EAAE,CAAC,MAAM,CAAC,eAAe,CAAC,KAAK,EAAE,MAAM,EAAE,MAAM,CAAC;YACtE,MAAM,EAAE,GAAG,EAAE,CAAC,MAAM,CAAC,MAAM,CAAC,KAAK,CAAC;SAClC,CAAC,CAAC;QACH,IAAI,MAAM,CAAC,SAAS;YAAE,OAAO;QAC7B,MAAM,WAAW,CAAC,GAAG,EAAE,MAAM,EAAE,MAAM,CAAC,KAAK,CAAC,CAAC;QAC7C,GAAG,CAAC,EAAE,CAAC,MAAM,CAAC,cAAc,KAAK,EAAE,CAAC,CAAC;IAAA,CACrC,CAAC;IAEF,EAAE,CAAC,eAAe,CAAC,OAAO,EAAE;QAC3B,WAAW,EAAE,gCAAgC;QAC7C,OAAO,EAAE,KAAK,EAAE,KAAK,EAAE,GAAG,EAAE,EAAE,CAAC;YAC9B,IAAI,GAAG,CAAC,IAAI,KAAK,KAAK,EAAE,CAAC;gBACxB,GAAG,CAAC,EAAE,CAAC,MAAM,CAAC,yCAAyC,EAAE,SAAS,CAAC,CAAC;gBACpE,OAAO;YACR,CAAC;YACD,MAAM,MAAM,GAAG,MAAM,gBAAgB,CAAC,GAAG,CAAC,CAAC;YAC3C,IAAI,CAAC,MAAM;gBAAE,OAAO;YACpB,MAAM,IAAA,mBAAW,EAAC,GAAG,EAAE,KAAK,EAAE,EAAE,EAAE,EAAE,CAAC;gBACpC,MAAM,WAAW,GAAG,KAAK,IAA2C,EAAE,CAAC;oBACtE,OAAO,IAAI,EAAE,CAAC;wBACb,IAAI,CAAC;4BACJ,OAAO,MAAM,WAAW,CAAC,GAAG,EAAE,MAAM,CAAC,CAAC;wBACvC,CAAC;wBAAC,OAAO,KAAK,EAAE,CAAC;4BAChB,IAAI,CAAC,MAAM,EAAE,CAAC,eAAe,CAAC,MAAM,CAAC,SAAS,EAAE,sBAAsB,CAAC,KAAK,CAAC,CAAC,CAAC,KAAK,OAAO,EAAE,CAAC;gCAC7F,OAAO,SAAS,CAAC;4BAClB,CAAC;wBACF,CAAC;oBACF,CAAC;gBAAA,CACD,CAAC;gBAEF,IAAI,OAAO,GAAG,MAAM,WAAW,EAAE,CAAC;gBAClC,IAAI,CAAC,OAAO;oBAAE,OAAO;gBACrB,OAAO,IAAI,EAAE,CAAC;oBACb,MAAM,MAAM,GAAG,MAAM,EAAE,CAAC,UAAU,CAAC,MAAM,CAAC,SAAS,EAAE,OAAO,CAAC,CAAC;oBAC9D,IAAI,MAAM,CAAC,IAAI,KAAK,OAAO;wBAAE,OAAO;oBACpC,IAAI,WAAoB,CAAC;oBACzB,IAAI,CAAC;wBACJ,IAAI,MAAM,CAAC,IAAI,KAAK,UAAU;4BAAE,MAAM,aAAa,CAAC,GAAG,EAAE,EAAE,EAAE,MAAM,CAAC,CAAC;6BAChE,IAAI,aAAa,CAAC,MAAM,CAAC,KAAK,CAAC;4BAAE,MAAM,WAAW,CAAC,GAAG,EAAE,EAAE,EAAE,MAAM,EAAE,MAAM,CAAC,KAAK,CAAC,CAAC;6BAClF,IAAI,MAAM,CAAC,KAAK,CAAC,MAAM,CAAC,KAAK,KAAK,UAAU;4BAChD,MAAM,SAAS,CAAC,GAAG,EAAE,EAAE,EAAE,MAAM,EAAE,OAAO,EAAE,MAAM,CAAC,KAAK,CAAC,CAAC;;4BACpD,GAAG,CAAC,EAAE,CAAC,MAAM,CAAC,GAAG,MAAM,CAAC,KAAK,CAAC,EAAE,OAAO,MAAM,CAAC,KAAK,CAAC,MAAM,CAAC,KAAK,EAAE,EAAE,SAAS,CAAC,CAAC;oBACrF,CAAC;oBAAC,OAAO,KAAK,EAAE,CAAC;wBAChB,WAAW,GAAG,KAAK,CAAC;oBACrB,CAAC;oBACD,MAAM,SAAS,GAAG,MAAM,WAAW,EAAE,CAAC;oBACtC,IAAI,CAAC,SAAS;wBAAE,OAAO;oBACvB,OAAO,GAAG,SAAS,CAAC;oBACpB,IAAI,WAAW,IAAI,CAAC,iBAAiB,CAAC,WAAW,CAAC,EAAE,CAAC;wBACpD,GAAG,CAAC,EAAE,CAAC,MAAM,CAAC,WAAW,YAAY,KAAK,CAAC,CAAC,CAAC,WAAW,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,WAAW,CAAC,EAAE,OAAO,CAAC,CAAC;oBAClG,CAAC;gBACF,CAAC;YAAA,CACD,CAAC,CAAC;QAAA,CACH;KACD,CAAC,CAAC;AAAA,CACH","sourcesContent":["import type { ExtensionAPI, ExtensionCommandContext } from \"../../core/extensions/types.ts\";\nimport { formatBytes, LlamaClient, type LlamaModelInfo, normalizeLlamaServerUrl } from \"./client.ts\";\nimport { findHuggingFaceToken, HuggingFaceClient } from \"./huggingface.ts\";\nimport { createLlamaProvider, LLAMA_PROVIDER_ID } from \"./provider.ts\";\nimport { type LlamaUi, runWithProgress, showLlamaUi } from \"./ui.ts\";\n\nfunction modelIsLoaded(model: LlamaModelInfo): boolean {\n\treturn model.status.value === \"loaded\" || model.status.value === \"sleeping\";\n}\n\nfunction isConnectionError(error: unknown): boolean {\n\tif (!(error instanceof Error)) return false;\n\tconst message = `${error.name} ${error.message}`.toLowerCase();\n\treturn message.includes(\"fetch failed\") || message.includes(\"timeout\") || message.includes(\"network\");\n}\n\nfunction connectionErrorMessage(error: unknown): string {\n\tif (isConnectionError(error)) return \"Could not connect to the server.\";\n\treturn error instanceof Error ? error.message : String(error);\n}\n\nfunction parseHuggingFaceModel(value: string): { repository: string; quantization?: string } {\n\tconst colon = value.indexOf(\":\", value.indexOf(\"/\") + 1);\n\treturn colon < 0\n\t\t? { repository: value }\n\t\t: { repository: value.slice(0, colon), quantization: value.slice(colon + 1) };\n}\n\nasync function configuredClient(ctx: ExtensionCommandContext): Promise<LlamaClient | undefined> {\n\tconst result = await ctx.modelRegistry.getProviderAuth(LLAMA_PROVIDER_ID);\n\tif (!result) {\n\t\tctx.ui.notify(`Configure llama.cpp with /login ${LLAMA_PROVIDER_ID}`, \"warning\");\n\t\treturn undefined;\n\t}\n\tconst configuredUrl = result.env?.LLAMA_BASE_URL;\n\tconst serverUrl = normalizeLlamaServerUrl(\n\t\ttypeof configuredUrl === \"string\" && configuredUrl ? configuredUrl : (result.auth.baseUrl ?? \"\"),\n\t);\n\treturn new LlamaClient(serverUrl, result.auth.apiKey);\n}\n\nexport default function llamaExtension(pi: ExtensionAPI): void {\n\tconst provider = createLlamaProvider();\n\tpi.registerProvider(provider.provider);\n\n\tconst syncCatalog = async (\n\t\tctx: ExtensionCommandContext,\n\t\tclient: LlamaClient,\n\t\tcatalog?: LlamaModelInfo[],\n\t): Promise<LlamaModelInfo[]> => {\n\t\tconst current = catalog ?? (await client.list());\n\t\tprovider.setCatalog(current, client.serverUrl);\n\t\tawait ctx.modelRegistry.refresh();\n\t\treturn current;\n\t};\n\n\tconst loadModel = async (\n\t\tctx: ExtensionCommandContext,\n\t\tui: LlamaUi,\n\t\tclient: LlamaClient,\n\t\tcatalog: LlamaModelInfo[],\n\t\ttarget: LlamaModelInfo,\n\t): Promise<void> => {\n\t\tconst loaded = catalog.filter((model) => model.id !== target.id && modelIsLoaded(model));\n\t\tlet replace = false;\n\t\tif (loaded.length > 0) {\n\t\t\tconst choice = await ui.select(`${loaded.length} model${loaded.length === 1 ? \" is\" : \"s are\"} loaded`, [\n\t\t\t\t\"Unload all and load\",\n\t\t\t\t\"Keep loaded and load\",\n\t\t\t\t\"Cancel\",\n\t\t\t]);\n\t\t\tif (!choice || choice === \"Cancel\") return;\n\t\t\treplace = choice === \"Unload all and load\";\n\t\t}\n\n\t\tconst restoreLoaded = async (): Promise<void> => {\n\t\t\tctx.ui.notify(\"Restoring previously loaded models\");\n\t\t\tfor (const model of loaded) await client.loadAndWait(model.id, () => {});\n\t\t\tawait syncCatalog(ctx, client);\n\t\t};\n\t\tif (replace) {\n\t\t\tfor (const model of loaded) await client.unloadAndWait(model.id);\n\t\t}\n\n\t\ttry {\n\t\t\tconst result = await runWithProgress(ui, {\n\t\t\t\ttitle: \"Loading model\",\n\t\t\t\tmodel: target.id,\n\t\t\t\tinitialMessage: \"Starting…\",\n\t\t\t\tcancelTitle: \"Stop loading?\",\n\t\t\t\tcancelMessage: target.id,\n\t\t\t\trun: (signal, update) => client.loadAndWait(target.id, update, signal),\n\t\t\t\tcancel: () => client.unload(target.id),\n\t\t\t});\n\t\t\tif (result.cancelled) {\n\t\t\t\tif (replace) await restoreLoaded();\n\t\t\t\treturn;\n\t\t\t}\n\t\t\tconst refreshed = await syncCatalog(ctx, client);\n\t\t\tconst loadedModel = refreshed.find((model) => model.id === target.id);\n\t\t\tctx.ui.notify(\n\t\t\t\tloadedModel?.status.value === \"loaded\" ? `Loaded ${target.id}` : `Load started for ${target.id}`,\n\t\t\t);\n\t\t} catch (error) {\n\t\t\tif (replace) {\n\t\t\t\ttry {\n\t\t\t\t\tawait restoreLoaded();\n\t\t\t\t} catch {\n\t\t\t\t\t// Preserve the original load error.\n\t\t\t\t}\n\t\t\t}\n\t\t\tthrow error;\n\t\t}\n\t};\n\n\tconst unloadModel = async (\n\t\tctx: ExtensionCommandContext,\n\t\tui: LlamaUi,\n\t\tclient: LlamaClient,\n\t\tmodel: LlamaModelInfo,\n\t): Promise<void> => {\n\t\tif (!(await ui.confirm(\"Unload model?\", model.id))) return;\n\t\tawait client.unloadAndWait(model.id);\n\t\tawait syncCatalog(ctx, client);\n\t\tctx.ui.notify(`Unloaded ${model.id}`);\n\t};\n\n\tconst downloadModel = async (ctx: ExtensionCommandContext, ui: LlamaUi, client: LlamaClient): Promise<void> => {\n\t\tconst huggingFace = new HuggingFaceClient(await findHuggingFaceToken());\n\t\tconst selected = await ui.searchModels((query, signal) => huggingFace.search(query, signal));\n\t\tif (!selected) return;\n\t\tconst parsed = parseHuggingFaceModel(selected);\n\t\tui.showStatus(\"Loading model details\", parsed.repository);\n\t\tconst details = await huggingFace.details(parsed.repository);\n\t\tif (details.gated) {\n\t\t\tconst approval = details.gated === \"manual\" ? \"Manual approval is required\" : \"Accept the access terms\";\n\t\t\tconst choice = await ui.select(\n\t\t\t\t`Hugging Face access required\\n${details.id}\\n\\n${approval} at:\\nhttps://huggingface.co/${details.id}\\n\\nThe llama.cpp server needs HF_TOKEN with access.`,\n\t\t\t\t[\"Continue\", \"Back\"],\n\t\t\t);\n\t\t\tif (choice !== \"Continue\") return;\n\t\t}\n\t\tlet quantization = parsed.quantization;\n\t\tif (!quantization && details.quantizations.length > 0) {\n\t\t\tconst options = details.quantizations.map((entry) => {\n\t\t\t\tconst detail = [\n\t\t\t\t\tentry.size === undefined ? undefined : formatBytes(entry.size),\n\t\t\t\t\tentry.name === \"Q4_K_M\" ? \"recommended\" : undefined,\n\t\t\t\t]\n\t\t\t\t\t.filter((value): value is string => Boolean(value))\n\t\t\t\t\t.join(\" · \");\n\t\t\t\treturn detail ? `${entry.name} · ${detail}` : entry.name;\n\t\t\t});\n\t\t\tconst choice = await ui.select(`Select quantization\\n${details.id}`, options);\n\t\t\tif (!choice) return;\n\t\t\tquantization = details.quantizations[options.indexOf(choice)]?.name;\n\t\t\tif (!quantization) return;\n\t\t}\n\t\tconst model = quantization ? `${details.id}:${quantization}` : details.id;\n\t\tconst result = await runWithProgress(ui, {\n\t\t\ttitle: \"Downloading model\",\n\t\t\tmodel,\n\t\t\tinitialMessage: \"Starting…\",\n\t\t\tcancelTitle: \"Stop download?\",\n\t\t\tcancelMessage: model,\n\t\t\trun: (signal, update) => client.downloadAndWait(model, update, signal),\n\t\t\tcancel: () => client.unload(model),\n\t\t});\n\t\tif (result.cancelled) return;\n\t\tawait syncCatalog(ctx, client, result.value);\n\t\tctx.ui.notify(`Downloaded ${model}`);\n\t};\n\n\tpi.registerCommand(\"llama\", {\n\t\tdescription: \"Manage llama.cpp router models\",\n\t\thandler: async (_args, ctx) => {\n\t\t\tif (ctx.mode !== \"tui\") {\n\t\t\t\tctx.ui.notify(\"/llama is available in interactive mode\", \"warning\");\n\t\t\t\treturn;\n\t\t\t}\n\t\t\tconst client = await configuredClient(ctx);\n\t\t\tif (!client) return;\n\t\t\tawait showLlamaUi(ctx, async (ui) => {\n\t\t\t\tconst readCatalog = async (): Promise<LlamaModelInfo[] | undefined> => {\n\t\t\t\t\twhile (true) {\n\t\t\t\t\t\ttry {\n\t\t\t\t\t\t\treturn await syncCatalog(ctx, client);\n\t\t\t\t\t\t} catch (error) {\n\t\t\t\t\t\t\tif ((await ui.connectionError(client.serverUrl, connectionErrorMessage(error))) === \"close\") {\n\t\t\t\t\t\t\t\treturn undefined;\n\t\t\t\t\t\t\t}\n\t\t\t\t\t\t}\n\t\t\t\t\t}\n\t\t\t\t};\n\n\t\t\t\tlet catalog = await readCatalog();\n\t\t\t\tif (!catalog) return;\n\t\t\t\twhile (true) {\n\t\t\t\t\tconst action = await ui.showModels(client.serverUrl, catalog);\n\t\t\t\t\tif (action.type === \"close\") return;\n\t\t\t\t\tlet actionError: unknown;\n\t\t\t\t\ttry {\n\t\t\t\t\t\tif (action.type === \"download\") await downloadModel(ctx, ui, client);\n\t\t\t\t\t\telse if (modelIsLoaded(action.model)) await unloadModel(ctx, ui, client, action.model);\n\t\t\t\t\t\telse if (action.model.status.value === \"unloaded\")\n\t\t\t\t\t\t\tawait loadModel(ctx, ui, client, catalog, action.model);\n\t\t\t\t\t\telse ctx.ui.notify(`${action.model.id} is ${action.model.status.value}`, \"warning\");\n\t\t\t\t\t} catch (error) {\n\t\t\t\t\t\tactionError = error;\n\t\t\t\t\t}\n\t\t\t\t\tconst refreshed = await readCatalog();\n\t\t\t\t\tif (!refreshed) return;\n\t\t\t\t\tcatalog = refreshed;\n\t\t\t\t\tif (actionError && !isConnectionError(actionError)) {\n\t\t\t\t\t\tctx.ui.notify(actionError instanceof Error ? actionError.message : String(actionError), \"error\");\n\t\t\t\t\t}\n\t\t\t\t}\n\t\t\t});\n\t\t},\n\t});\n}\n"]}
@@ -1,10 +0,0 @@
1
- import type { Provider } from "@earendil-works/pi-ai";
2
- import { type LlamaModelInfo } from "./client.ts";
3
- export declare const LLAMA_PROVIDER_ID = "llama.cpp";
4
- export declare const DEFAULT_LLAMA_SERVER_URL = "http://127.0.0.1:8080";
5
- export interface LlamaProviderController {
6
- provider: Provider<"openai-completions">;
7
- setCatalog(models: readonly LlamaModelInfo[], serverUrl: string): void;
8
- }
9
- export declare function createLlamaProvider(): LlamaProviderController;
10
- //# sourceMappingURL=provider.d.ts.map
@@ -1 +0,0 @@
1
- {"version":3,"file":"provider.d.ts","sourceRoot":"","sources":["../../../src/extensions/llama/provider.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAKX,QAAQ,EAGR,MAAM,uBAAuB,CAAC;AAE/B,OAAO,EAAe,KAAK,cAAc,EAA8C,MAAM,aAAa,CAAC;AAE3G,eAAO,MAAM,iBAAiB,cAAc,CAAC;AAC7C,eAAO,MAAM,wBAAwB,0BAA0B,CAAC;AAyChE,MAAM,WAAW,uBAAuB;IACvC,QAAQ,EAAE,QAAQ,CAAC,oBAAoB,CAAC,CAAC;IACzC,UAAU,CAAC,MAAM,EAAE,SAAS,cAAc,EAAE,EAAE,SAAS,EAAE,MAAM,GAAG,IAAI,CAAC;CACvE;AAED,wBAAgB,mBAAmB,IAAI,uBAAuB,CAmE7D","sourcesContent":["import type {\n\tApiKeyCredential,\n\tAuthContext,\n\tAuthResult,\n\tModel,\n\tProvider,\n\tProviderStreamOptions,\n\tRefreshModelsContext,\n} from \"@earendil-works/pi-ai\";\nimport { stream, streamSimple } from \"@earendil-works/pi-ai/compat\";\nimport { LlamaClient, type LlamaModelInfo, llamaInferenceUrl, normalizeLlamaServerUrl } from \"./client.ts\";\n\nexport const LLAMA_PROVIDER_ID = \"llama.cpp\";\nexport const DEFAULT_LLAMA_SERVER_URL = \"http://127.0.0.1:8080\";\nconst DEFAULT_MAX_TOKENS = 16384;\n\nfunction credentialServerUrl(credential: ApiKeyCredential | undefined): string | undefined {\n\tconst value = credential?.env?.LLAMA_BASE_URL;\n\treturn typeof value === \"string\" && value.trim() ? normalizeLlamaServerUrl(value) : undefined;\n}\n\nasync function resolveServerUrl(\n\tctx: AuthContext,\n\tcredential: ApiKeyCredential | undefined,\n): Promise<string | undefined> {\n\tconst configured = credentialServerUrl(credential) ?? (await ctx.env(\"LLAMA_BASE_URL\"))?.trim();\n\treturn configured ? normalizeLlamaServerUrl(configured) : undefined;\n}\n\nfunction toPiModel(model: LlamaModelInfo, serverUrl: string): Model<\"openai-completions\"> {\n\tconst reportedContextWindow = model.meta?.n_ctx ?? model.meta?.n_ctx_train;\n\tconst contextWindow = reportedContextWindow && reportedContextWindow > 0 ? reportedContextWindow : 128000;\n\treturn {\n\t\tid: model.id,\n\t\tname: model.id,\n\t\tapi: \"openai-completions\",\n\t\tprovider: LLAMA_PROVIDER_ID,\n\t\tbaseUrl: llamaInferenceUrl(serverUrl),\n\t\treasoning: false,\n\t\tinput: model.architecture?.input_modalities?.includes(\"image\") ? [\"text\", \"image\"] : [\"text\"],\n\t\tcost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 },\n\t\tcontextWindow,\n\t\tmaxTokens: Math.min(DEFAULT_MAX_TOKENS, contextWindow),\n\t\tcompat: {\n\t\t\tsupportsStore: false,\n\t\t\tsupportsDeveloperRole: false,\n\t\t\tsupportsReasoningEffort: false,\n\t\t\tsupportsUsageInStreaming: false,\n\t\t\tsupportsStrictMode: false,\n\t\t\tmaxTokensField: \"max_tokens\",\n\t\t},\n\t};\n}\n\nexport interface LlamaProviderController {\n\tprovider: Provider<\"openai-completions\">;\n\tsetCatalog(models: readonly LlamaModelInfo[], serverUrl: string): void;\n}\n\nexport function createLlamaProvider(): LlamaProviderController {\n\tlet models: readonly Model<\"openai-completions\">[] = [];\n\n\tconst setCatalog = (catalog: readonly LlamaModelInfo[], serverUrl: string): void => {\n\t\tmodels = catalog.filter((model) => model.status.value === \"loaded\").map((model) => toPiModel(model, serverUrl));\n\t};\n\n\tconst provider: Provider<\"openai-completions\"> = {\n\t\tid: LLAMA_PROVIDER_ID,\n\t\tname: \"llama.cpp\",\n\t\tbaseUrl: llamaInferenceUrl(DEFAULT_LLAMA_SERVER_URL),\n\t\tauth: {\n\t\t\tapiKey: {\n\t\t\t\tname: \"llama.cpp server\",\n\t\t\t\tlogin: async (interaction): Promise<ApiKeyCredential> => {\n\t\t\t\t\tconst enteredUrl = await interaction.prompt({\n\t\t\t\t\t\ttype: \"text\",\n\t\t\t\t\t\tmessage: \"llama.cpp server URL\",\n\t\t\t\t\t\tplaceholder: process.env.LLAMA_BASE_URL ?? DEFAULT_LLAMA_SERVER_URL,\n\t\t\t\t\t});\n\t\t\t\t\tconst serverUrl = normalizeLlamaServerUrl(\n\t\t\t\t\t\tenteredUrl.trim() || process.env.LLAMA_BASE_URL || DEFAULT_LLAMA_SERVER_URL,\n\t\t\t\t\t);\n\t\t\t\t\tconst apiKey = (\n\t\t\t\t\t\tawait interaction.prompt({\n\t\t\t\t\t\t\ttype: \"secret\",\n\t\t\t\t\t\t\tmessage: \"API key (optional)\",\n\t\t\t\t\t\t})\n\t\t\t\t\t).trim();\n\t\t\t\t\tawait new LlamaClient(serverUrl, apiKey || undefined).list({ signal: interaction.signal });\n\t\t\t\t\treturn {\n\t\t\t\t\t\ttype: \"api_key\",\n\t\t\t\t\t\tkey: apiKey || undefined,\n\t\t\t\t\t\tenv: { LLAMA_BASE_URL: serverUrl },\n\t\t\t\t\t};\n\t\t\t\t},\n\t\t\t\tcheck: async ({ ctx, credential }) => {\n\t\t\t\t\tconst serverUrl = await resolveServerUrl(ctx, credential);\n\t\t\t\t\treturn serverUrl\n\t\t\t\t\t\t? { type: \"api_key\", source: credential ? \"stored credential\" : \"LLAMA_BASE_URL\" }\n\t\t\t\t\t\t: undefined;\n\t\t\t\t},\n\t\t\t\tresolve: async ({ ctx, credential }): Promise<AuthResult | undefined> => {\n\t\t\t\t\tconst serverUrl = await resolveServerUrl(ctx, credential);\n\t\t\t\t\tif (!serverUrl) return undefined;\n\t\t\t\t\tconst apiKey = credential?.key ?? (await ctx.env(\"LLAMA_API_KEY\")) ?? \"local\";\n\t\t\t\t\treturn {\n\t\t\t\t\t\tauth: { apiKey, baseUrl: llamaInferenceUrl(serverUrl) },\n\t\t\t\t\t\tenv: { ...credential?.env, LLAMA_BASE_URL: serverUrl },\n\t\t\t\t\t\tsource: credential ? \"stored credential\" : \"LLAMA_BASE_URL\",\n\t\t\t\t\t};\n\t\t\t\t},\n\t\t\t},\n\t\t},\n\t\tgetModels: () => models,\n\t\trefreshModels: async (context: RefreshModelsContext): Promise<void> => {\n\t\t\tif (!context.allowNetwork || context.signal?.aborted || context.credential?.type !== \"api_key\") return;\n\t\t\tconst serverUrl = credentialServerUrl(context.credential);\n\t\t\tif (!serverUrl) return;\n\t\t\tconst catalog = await new LlamaClient(serverUrl, context.credential.key).list({ signal: context.signal });\n\t\t\tsetCatalog(catalog, serverUrl);\n\t\t},\n\t\tstream: (model, context, options) => stream(model, context, options as ProviderStreamOptions | undefined),\n\t\tstreamSimple: (model, context, options) => streamSimple(model, context, options),\n\t};\n\n\treturn { provider, setCatalog };\n}\n"]}