[BugFix] Fix potential cuda-graph IMA

[LucasWilkinson]

Essential Elements of an Effective PR Description Checklist

• The purpose of the PR, such as "Fix some issue (link existing issues this PR will resolve)".
• The test plan, such as providing test command.
• The test results, such as pasting the results comparison before and after, or e2e results
• (Optional) The necessary documentation update, such as updating supported_models.md and examples for a new model.

Purpose

Cuda-graph padding happens after prepare inputs; it's safer to -1 fill here (and closer behavior to pre #20466 ). No errors reported yet this is just preventative

Test Plan

lm_eval

Test Result

VLLM_ATTENTION_BACKEND=TRITON_ATTN_VLLM_V1  lm_eval   --model vllm   --model_args '{
    "pretrained": "meta-llama/Meta-Llama-3-8B-Instruct",
    "compilation_config": { "full_cuda_graph": true }
  }'   --tasks gsm8k   --batch_size auto 
...
|Tasks|Version|     Filter     |n-shot|  Metric   |   |Value |   |Stderr|
|-----|------:|----------------|-----:|-----------|---|-----:|---|-----:|
|gsm8k|      3|flexible-extract|     5|exact_match|↑  |0.7559|±  |0.0118|
|     |       |strict-match    |     5|exact_match|↑  |0.7574|±  |0.0118|

(Optional) Documentation Update

[github-actions]

👋 Hi! Thank you for contributing to the vLLM project.

💬 Join our developer Slack at https://slack.vllm.ai to discuss your PR in #pr-reviews, coordinate on features in #feat- channels, or join special interest groups in #sig- channels.

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 either: Add ready label to the PR or enable auto-merge.

🚀

[gemini-code-assist]

Code Review

This PR addresses a bug in full CUDA graph mode related to incorrect padding of slot_mapping. The removal of the ineffective __post_init__ method is correct. However, the new logic targets an incorrect tensor. The provided critical comment suggests a fix to target the correct slot_mapping tensors within self.input_batch.block_table.

[mgoin]

Seems reasonable to me, I guess there shouldn't be a performance concern aside from this being potentially moved out of the cudagraph?

[LucasWilkinson]

Seems reasonable to me, I guess there shouldn't be a performance concern aside from this being potentially moved out of the cudagraph?

Basically we just fill more during input prep; when it was in CommonAttentionMetadata it would only fill the already sliced tensor (we want to fill the whole thing incase we pad for cuda-graphs)