[Do Not Merge] - LoRA V1 Reference PR

[varun-sundar-rabindranath]

LoRA works end-to-end with this PR. However, this PR introduces a lot of changes that (some unnecessary) need to split up into smaller PRs. Putting up this PR as a reference for sub PRs.

Benchmarks:
Machine - 1xA100

Command:

VLLM_USE_V1="1" python3 benchmarks/benchmark_throughput.py --model  meta-llama/Llama-2-7b-hf --backend vllm   --dataset ./ShareGPT_V3_unfiltered_cleaned_split.json --num-prompts 1000  --max-loras 4  --max-lora-rank 8 --enable-lora --lora-path "yard1/llama-2-7b-sql-lora-test" -O 3

Throughput: 10.96 requests/s, 5272.77 total tokens/s, 2577.68 output tokens/s

python3 benchmarks/benchmark_throughput.py --model  meta-llama/Llama-2-7b-hf --backend vllm   --dataset ./ShareGPT_V3_unfiltered_cleaned_split.json --num-prompts 1000  --max-loras 4  --max-lora-rank 8 --enable-lora --lora-path "yard1/llama-2-7b-sql-lora-test" --num-scheduler-steps 8

Throughput: 9.60 requests/s, 4617.92 total tokens/s, 2257.54 output tokens/s

Plan for PR split:

• Base PR (changes to gpu_model_runner.py, v1 scheduler, and prefix caching) - [V1] LoRA Support #10957
• Changes to support torch.compile for LoRA - [Kernel] LoRA - Enable CUDAGraphs for V1 #14626
• Add LoRA kernels for V1, Autotune LoRA kernels for V1 - [V1] LoRA - Add triton kernels for V1 #13096
• Add and Test add_lora - [V1] LoRA - Enable Serving Usecase #12883
• Add and Test remove/pin LoRA functions - [Core] LoRA V1 - Add add/pin/list/remove_lora functions  #13705

[github-actions]

👋 Hi! Thank you for contributing to the vLLM project.
Just a reminder: PRs would not trigger full CI run by default. Instead, it would only run fastcheck CI which starts running only a small and essential subset of CI tests to quickly catch errors. You can run other CI tests on top of those by going to your fastcheck build on Buildkite UI (linked in the PR checks section) and unblock them. If you do not have permission to unblock, ping simon-mo or khluu to add you in our Buildkite org.

Once the PR is approved and ready to go, your PR reviewer(s) can run CI to test the changes comprehensively before merging.

To run CI, PR reviewers can do one of these:

• Add ready label to the PR
• Enable auto-merge.

🚀

[mergify]

This pull request has merge conflicts that must be resolved before it can be
merged. Please rebase the PR, @varun-sundar-rabindranath.

https://docs.github.com/en/pull-requests/collaborating-with-pull-requests/working-with-forks/syncing-a-fork

[mergify]

This pull request has merge conflicts that must be resolved before it can be
merged. Please rebase the PR, @varun-sundar-rabindranath.

https://docs.github.com/en/pull-requests/collaborating-with-pull-requests/working-with-forks/syncing-a-fork

[varun-sundar-rabindranath]

All functionality, except the triton configs, have landed on main.