Misc. bug: Tokens in top_probs / top_logprobs are missing whitespace

[CyberShadow]

Name and Version

ggml_cuda_init: GGML_CUDA_FORCE_MMQ: no
ggml_cuda_init: GGML_CUDA_FORCE_CUBLAS: no
ggml_cuda_init: found 1 CUDA devices:
Device 0: NVIDIA GeForce RTX 3090 Ti, compute capability 8.6, VMM: yes
version: 0 (unknown)
built with gcc (GCC) 13.3.0 for x86_64-unknown-linux-gnu

(actually version 4552, built with Nix)

Operating systems

Linux

Which llama.cpp modules do you know to be affected?

llama-server

Command line

$ bin/llama-server -m ../../../wizardcoder-python-34b-v1.0.Q5_K_M.gguf -ngl 9999
...

$ curl -fsS \
    --url http://127.0.0.1:8080/completion \
    --header "Content-Type: application/json" \
    --data '{"prompt": "Hello","n_predict": 1, "n_probs": 10, "temperature":0}' | jq .
{
  ...
  "completion_probabilities": [
    {
      "id": 2897,
      "token": " os",                      <---------- whitespace OK
      "bytes": [
        32,                                <---------- whitespace OK
        111,
        115
      ],
      "logprob": -2.0750603675842285,
      "top_logprobs": [
        {
          "id": 2897,
          "token": "os",                   <---------- whitespace missing
          "bytes": [
            111,                           <---------- whitespace missing
            115
          ],
          "logprob": -2.0750603675842285
        },

Problem description & steps to reproduce

As above, doesn't seem to depend on the model

First Bad Commit

57bb2c4

Relevant log output

[ngxson]

I don't have time to debug this right now, would appreciate if someone can take a look.

@danbev recently you've been spending time on llama-server, would you mind to have a look on this? Thanks.