[do not merge][CB] requesting only one token via index (fms api change)

[yannicks1]

[do not merge][CB] requesting only one token via index (fms api change)

To minimize data transfer for CB, we only want to request the last prompt tokens logits instead of the the entire prompts logits.
Therefore the flag only_last_token: bool will be replaced by the argument index: int in the fms forward api. When passing and index i, fms will return the logits for the token at i-th position only. A draft implementation of this in fms can be found here.

Note: we only request the last token logits for static batching by default, as no right padding is ever required there.

changes:

• install fms feature branch for testing
• passing index instead of only_last_token
• set number of right pads for decode to 0

solves #254

[github-actions]

👋 Hi! Thank you for contributing to vLLM support on Spyre.
Just a reminder: Make sure that your code passes all the linting checks, otherwise your PR won't be able to be merged. To do so, first install the linting requirements, then run format.sh and commit the changes. This can be done with uv directly:

uv sync --frozen --group lint --active --inexact

Or this can be done with pip:

uv pip compile --group lint > requirements-lint.txt
pip install -r requirements-lint.txt
bash format.sh

Now you are good to go 🚀