fix streaming test

[radames]

Not sure if this is due to the non-deterministic of text generation, however the space before the " one" keep making the test fail. So I've added the index to add the space WDYT @vvmnnnkv

Array(5) [
  0: Object {
  token: Object {id: 80, text: "one", logprob: -0.75390625, special: false}
  generated_text: null
  details: null
}
  1: Object {
  token: Object {id: 192, text: " two", logprob: -0.01940918, special: false}
  generated_text: null
  details: null
}
  2: Object {
  token: Object {id: 386, text: " three", logprob: -0.015197754, special: false}
  generated_text: null
  details: null
}
  3: Object {
  token: Object {id: 662, text: " four", logprob: -0.01940918, special: false}
  generated_text: null
  details: null
}
  4: Object {
  token: Object {id: 1, text: "</s>", logprob: -0.030883789, special: true}
  generated_text: "one two three four"
  details: null
}
]

[coyotte508]

It seems the generated response changed @radames

we can ignore spaces imo in the token, eg text: expect.stringContaining(tokenText) to be more robust

[vvmnnnkv]

It might be due to update of text-generation-inference to v0.5.0, noticed it has changes related to token decoding in huggingface/text-generation-inference#144.

[vvmnnnkv]

It's funny that tapes.json would update with random outputs each time :)