Skip to content

[Bug] The neural network module does not work on an Intel iMac running Sequoia 15.7.7 #755

Description

@FuJacob

Bug Report

Description

The program runs, the model loads, but the text substitution doesn't work at all. I'm using the latest iMac on Intel CPU (8x3.8GHz, 128Gb RAM). Llama.cpp in cli-mode works perfectly. Voice transcribbing apps like Handy works lightning fast too. I tried to perform Tests in CotabbyInference package and see: [CotabbyInferenceTests.LlamaMiddlewareTests testInterleavedMultiSequenceSampling] : XCTAssertEqual failed: ("[108, 607, 236761, 1023, 999, 108, 237314, 200]") is not equal to ("[108, 564, 236858, 236743, 858, 18765, 138, 236800]") - Greedy sampling with identical prompts should match across sequences. All other tests pass. Seems that the problem in CotabbyInference package.

Environment

  • Cotabby version: 0.6.0-beta
  • macOS version: 15.7.7
  • Mac model: iMac20,1
  • Chip: Intel(R) Core(TM) i7-10700K CPU @ 3.80GHz
  • Memory: 128 GB

Submitted via tabby feedback form

Metadata

Metadata

Assignees

Labels

area:perfCrashes, hangs, CPU pinning, slow responsebugSomething isn't working

Projects

Status
Backlog

Milestone

No milestone

Relationships

None yet

Development

No branches or pull requests

Issue actions