Stop not always working, causing duplicate output

[nico202]

To upvote this issue, give it a thumbs up. See this list for the most upvoted issues.

Please avoid AI slops, be concise, and focus on what matters for this issue.

Describe the bug

Sometimes stop does not actually stop the generation, and I end up with multiple overlapping answers.
I'm using eca-emacs, not sure where this bug belongs.

Example output:
Ervixplorce, e

Per cl'orompchestletratorare le spe agecgiiuficnge l'he, ho eventancora o alcuniall chiara coimenti da connec setassartoi `da_c:ategor

**i1zza. Tripi die`

eventi2. nFellase di caategorizz caziodonea:

• Me**ssa:ggi L'LLMut viene inenterte (da Terogalto per cegram, ompMletare atrix,inform azioEmnail) i ma→ dopo vencrificaanti (prfiirma?
• Torità, deriggera evedline, tnt pier proatmetiout) vie atggià orn(es. "mare il daeetitab

(two instances are writing their own output simultaneously. If I press STOP, one of the two stops but the other continue)

To Reproduce
I don't have a simple sequence of steps to reproduce. It just happens randomly if I interrupt the generation.

Expected behavior

Stop should always stop the agent.

Additional context
I'm using local llama.cpp

"providers": {
    "llama": {
      "api": "openai-chat",
      "url": "http://localhost:8080/v1",
      "fetchModels": true
    }
  }

[ericdallo]

I believe this happens when you stop and sned a prompt too quickly, but I need to repro to be able to fix it

[nico202]

I don't think it requires the actions to be fast. Sometimes after pressing stop I still see llama-server processing the prompt. Only when the processing finishes, eca's postRequest hook triggers. I need to check the code, but maybe it has to do with the fact that the request has not received any token yet?