VLLM: boolean flags and constraints

[podkidyshev]

Summary

For VLLM:

• fixes wrong way of handling boolean flags;
• Adds constraints checks.

Test Plan

• Automated CI

Additional Notes

[coderabbitai]

📝 Walkthrough

Walkthrough

Extracted GPU-selection helpers and changed two-node disaggregated GPU selection; added boolean-aware vLLM CLI flag serialization; implemented vLLM parallelism constraint checks validating prefill/decode roles; updated tests for flag serialization and constraint scenarios.

Changes

Cohort / File(s)	Summary
GPU ID selection & serve-arg serialization src/cloudai/workloads/common/llm_serving.py	Added calculate_prefill_gpu_ids(...) and calculate_decode_gpu_ids(...); refactored LLMServingSlurmCommandGenStrategy.prefill_gpu_ids and .decode_gpu_ids to use them; added LLMServingArgs.serialize_serve_arg(...). Modified two-node disaggregated automatic GPU selection to return full GPU list when num_nodes == 2.
vLLM CLI flags & constraint logic src/cloudai/workloads/vllm/vllm.py	Added VllmArgs.serialize_serve_arg(...) to emit kebab-case boolean flags as --flag / --no-flag; introduced _validate_vllm_parallelism_constraints(...) and VllmTestDefinition.constraint_check(...) to validate tp/pp/dp against available GPUs and enforce backend compatibility, using new GPU helpers for prefill/decode checks.
Tests: vLLM serve args & constraints tests/workloads/vllm/test_command_gen_strategy_slurm.py, tests/workloads/vllm/test_workload.py	Added tests verifying boolean flag serialization in generated vLLM serve commands and multiple constraint_check scenarios (GPU capacity, backend compatibility, disaggregated allocations, two-node behavior).

Estimated code review effort

🎯 3 (Moderate) | ⏱️ ~25 minutes

Poem

🐇 I hop where GPUs line the trail,

Prefill and decode tell the tale.
Flags flip true, or no — with glee,
Constraints checked for harmony.
A little rabbit claps — hooray for thee!

🚥 Pre-merge checks | ✅ 2

✅ Passed checks (2 passed)

Check name	Status	Explanation
Title check	✅ Passed	The title clearly and concisely summarizes the main changes: fixing boolean flag handling and adding constraint checks for VLLM.
Description check	✅ Passed	The description accurately covers the two main objectives: boolean flags fix and constraints checks, with appropriate testing plan.

_{✏️ Tip: You can configure your own custom pre-merge checks in the settings.}

✨ Finishing Touches

📝 Generate docstrings

• Create stacked PR
• Commit on current branch

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch ipod/vllm-bool-and-constraints

---

_{Comment @coderabbitai help to get the list of available commands and usage tips.}

[coderabbitai]

Caution

Failed to replace (edit) comment. This is likely due to insufficient permissions or the comment being deleted.

Error details

{"name":"HttpError","status":401,"request":{"method":"PATCH","url":"https://api.github.com/repos/NVIDIA/cloudai/issues/comments/4163563183","headers":{"accept":"application/vnd.github.v3+json","user-agent":"octokit.js/0.0.0-development octokit-core.js/7.0.6 Node.js/24","authorization":"token [REDACTED]","content-type":"application/json; charset=utf-8"},"body":{"body":"<!-- This is an auto-generated comment: summarize by coderabbit.ai -->\nNo actionable comments were generated in the recent review. 🎉\n\n<details>\n<summary>ℹ️ Recent review info</summary>\n\n<details>\n<summary>⚙️ Run configuration</summary>\n\n**Configuration used**: Path: .coderabbit.yaml\n\n**Review profile**: ASSERTIVE\n\n**Plan**: Pro\n\n**Run ID**: `12a226dc-052e-40b7-8e07-db938218d3bf`\n\n</details>\n\n<details>\n<summary>📥 Commits</summary>\n\nReviewing files that changed from the base of the PR and between 3ad86b8bcd335c55275cd0ab9eda8e911234e9ff and 1df854bec49677feac16bd77b8b3ea739f760570.\n\n</details>\n\n<details>\n<summary>📒 Files selected for processing (4)</summary>\n\n* `src/cloudai/workloads/common/llm_serving.py`\n* `src/cloudai/workloads/vllm/vllm.py`\n* `tests/workloads/vllm/test_command_gen_strategy_slurm.py`\n* `tests/workloads/vllm/test_workload.py`\n\n</details>\n\n</details>\n\n---\n\n\n<!-- walkthrough_start -->\n\n<details>\n<summary>📝 Walkthrough</summary>\n\n## Walkthrough\n\nGPU-ID calculation logic is refactored into centralized helper functions for prefill and decode operations. vLLM workload gains CLI argument generation via serve_args property and comprehensive parallelism constraint validation covering tensor/pipeline/data parallel limits and backend-specific rules. Tests verify boolean flag conversion and multi-pattern constraint checking.\n\n## Changes\n\n|Cohort / File(s)|Summary|\n|---|---|\n|**GPU ID Calculation Helpers** <br> `src/cloudai/workloads/common/llm_serving.py`|Added `calculate_prefill_gpu_ids()` and `calculate_decode_gpu_ids()` helper functions; refactored `LLMServingSlurmCommandGenStrategy.prefill_gpu_ids` and `.decode_gpu_ids` properties to delegate to these helpers instead of inline logic.|\n|**vLLM Constraint Validation & Serve Args** <br> `src/cloudai/workloads/vllm/vllm.py`|Added `VllmArgs.serve_args` property to generate vLLM CLI arguments with kebab-case keys and boolean flag handling; introduced `VllmTestDefinition.constraint_check()` method and `_validate_vllm_parallelism_constraints()` static method to enforce tensor/pipeline/data-parallel limits and backend-specific rules; updated imports for constraint validation utilities.|\n|**vLLM Boolean Flag & Constraint Tests** <br> `tests/workloads/vllm/test_command_gen_strategy_slurm.py`, `tests/workloads/vllm/test_workload.py`|Added test for boolean flag conversion (`--enable-expert-parallel`); added test assertions for `serve_args` property with boolean-to-flag mapping; added constraint-check test cases covering GPU exhaustion, backend rule violations, and disaggregated role alignment scenarios.|\n\n## Estimated code review effort\n\n🎯 4 (Complex) | ⏱️ ~50 minutes\n\n## Poem\n\n> 🐰 *GPU IDs organized, constraints now aligned,*  \n> *vLLM commands flourish with CLI redesigned,*  \n> *Prefill and decode dance in harmony true,*  \n> *Parallelism checks keep the system askew-free too!* ✨\n\n</details>\n\n<!-- walkthrough_end -->\n\n\n<!-- pre_merge_checks_walkthrough_start -->\n\n<details>\n<summary>🚥 Pre-merge checks | ✅ 2</summary>\n\n<details>\n<summary>✅ Passed checks (2 passed)</summary>\n\n|     Check name    | Status   | Explanation                                                                                                                                                 |\n| :---------------: | :------- | :---------------------------------------------------------------------------------------------------------------------------------------------------------- |\n|    Title check    | ✅ Passed | The title 'VLLM: boolean flags and constraints' clearly identifies the main changes: fixing boolean flag handling and adding constraint validation in VLLM. |\n| Description check | ✅ Passed | The description directly addresses the changeset, explaining the two main objectives: fixing boolean flag handling and adding constraint checks in VLLM.    |\n\n</details>\n\n<sub>✏️ Tip: You can configure your own custom pre-merge checks in the settings.</sub>\n\n</details>\n\n<!-- pre_merge_checks_walkthrough_end -->\n\n<!-- finishing_touch_checkbox_start -->\n\n<details>\n<summary>✨ Finishing Touches</summary>\n\n<details>\n<summary>📝 Generate docstrings</summary>\n\n- [ ] <!-- {\"checkboxId\": \"7962f53c-55bc-4827-bfbf-6a18da830691\"} --> Create stacked PR\n- [ ] <!-- {\"checkboxId\": \"3e1879ae-f29b-4d0d-8e06-d12b7ba33d98\"} --> Commit on current branch\n\n</details>\n<details>\n<summary>🧪 Generate unit tests (beta)</summary>\n\n- [ ] <!-- {\"checkboxId\": \"f47ac10b-58cc-4372-a567-0e02b2c3d479\", \"radioGroupId\": \"utg-output-choice-group-unknown_comment_id\"} -->   Create PR with unit tests\n- [ ] <!-- {\"checkboxId\": \"6ba7b810-9dad-11d1-80b4-00c04fd430c8\", \"radioGroupId\": \"utg-output-choice-group-unknown_comment_id\"} -->   Commit unit tests in branch `ipod/vllm-bool-and-constraints`\n\n</details>\n\n</details>\n\n<!-- finishing_touch_checkbox_end -->\n\n<!-- tips_start -->\n\n---\n\n\n\n<sub>Comment `@coderabbitai help` to get the list of available commands and usage tips.</sub>\n\n<!-- tips_end -->\n\n<!-- internal state start -->\n\n\n<!-- DwQgtGAEAqAWCWBnSTIEMB26CuAXA9mAOYCmGJATmriQCaQDG+Ats2bgFyQAOFk+AIwBWJBrngA3EsgEBPRvlqU0AgfFwA6NPEgQAfACgjoCEYDEZyAAUASpETZWaCrKPR1AGxJcAagBk/AFkuAXx8L0xIADMPNCJkTHomDERcKngMXGRIAwA5RwFKLgAOAFYAdkgcgFUbPy5YXFxuRA4AejaidVhsAQ0mZjbcnwBJABERgEE2hg98bFptNu5sDw82ssqaxCKeRQBreFpZRFgSCSqDAGV5igYSSAEqDAZYLnhuRTaJNeYwUPCYESYGSqXSmWQgCTCGDOUi4R7PV5cZjaLA5K64ajYVr8bhkS42c7wEgAd0oOMQFEk1EQmH2VNOGEwABpICIolEMBIjvA0JcAMIUEjUOjoTiQABMAAYJQA2MBSgDMYEVAEZoKrShxFbLtRKAFpGMbSBhU7jifBYBxOFwcAxQABi8AAHtIUC98BQhWJILBEh4MkR+FFHmEIlgYnFkBlIP4ghpLlBJrRaMhXqJ9sgop6FCk0qisu7YwFAgmcgZoNJ4dxYhguJM8CwRfR+SMy0Zcvh0Cn1PBLWgPJAMPgaMhePhuUpaBojPpjOAoGR6PgQ2hG8QyMoaEkWGxMlxePxhKJxFIZPImEoqKp1FodHOTFA4KhUJE1wQN+QqNuFKx2FwqBJexHBRFxHgvRRlBvTRtF0MBDHnUwDEpBgZjmBYlhJT19jmNBUxmXdLTaX4AH0dgobkMCIDRuFkO0ACJGIMCxIEmEZPy3UVrVA+QV0YP0qOkAxkynSBcCwodSV9EgPDxPgomwF4LRSVkAAMGAHBhVhFEjeBIKJ4DWEiiBWEijkQAAKDRrIASlU9AMHodTNO0mgSKUS8SGM0zzKs2zVNZAhGHYKgAwALweABxKxqjAcZ7Bkk8+ywOYugYaIcz0gy1gc+gPMghNqm4RYf1UksrkoSiiCuDxsAoZh+V3RJIrIDFvxIIhZBooVso8bzsDM1N7MSSBVI0fKlH6wbEHs8c5PEN0go0nLcDOHZpNk8l3VSYVlxDDIA3IegnkwV5A0gVL4AYBM4AecggKywzB2WrTYmUyAhVwOqUlGgBtABdeySTOLB1OYWgSNhRBuv0p77NQRSdlwVluGcHZkBIZ0ayuzx5HHSdRVUx61g0EyBvMoGQZ4IUdkyVkRzOCgSSQB4r0kN1xmQbk+VUgc+rJ6a/I0OzcoSrwxGQVaHgMihUl9AcQ2B/EKEUjBzstSBmE9B5VsicSu2HJRoZgM5JKAiaHhe1zkvseBmEM5wPDxtG3Ux7GGHUUaLdJnyhsgJWsD02nkf4KWmZZyA2bPSBOcgbnRr5qbfOs4X7IEPBIGxN0LbARA8Q9gz0sKP1uU9d4QyJnq4ZQZBEZIeFPc+77JdNhScui6oLqQEP4BDDX9aHSDo3hJuKB+vnI8odn6E5+mw+Z9bPZ2cXCylhLknoP0PCiGcjGYyxJg8Ghv2SyWuzXjzYhPy1kD4t3PR/TLegDdL2F7ISkxTUV7uiRSxFPu0VQoDOQ8K9HSxN+a+0srgJQUQuDlUqoGSsqRjQGTVspX6CCKKBn5ODSYFB4jQH+qyDAjgSKG2kO8Om9gTg0GYP1RAulKDkMglQ+EAAfSAnZyA2V0HoLuqRfoZFwIDAwQDRpWzel5C2SdUwWRgfpeBQQKrYKosg3AqCMi9ktJg5RiCqK4NoPgwhxChxkIoTiYRrJEC0JIPQsmjC5IsKUGwyAnDuEkF4fBARuAhGZFEVAIqJVRRzUoLgeQdsax2PYNQU+kALKIHgEQJkX0aaZxeAJUgtAbKAN0KNLBVUap1Qak1RyLUMBtRFJ1GGvVZE4gDII4RgNB7m0SkQEUZ8JEuSkbpKuRkBbJ38mIvJZU9GqOqrVeqjUnBlNavmGg1SZEDNTFwBpvimn2WHK0rw7TRxiS7CAsBbkllQKFnZIwfgMhuleJgLJXAADUOo2hgFVBKIwABRVIdtmwKCUB9IkUl9LZgoOKAAEkk2ABhGL0VnEhFCaF5iLHgG0LCFAcL4Dwogb4vxsUeGYDROiUKmIsTYhxdq9BuLOF4iGG5glECzlYl/JyPhfjGOhuRKQkMCEzX2ZAUgX4RRxxLJAfkfgRjoAIY4dgWYKAsFGlrJQfVaCOG4GcgKYlniIGBfbKikB9gkBOLy1SEB9UCBUCCNAOx1V2PUOIXVAJwzRFiPEdAyBjVgDICoLwYBgD4HNHoVSbR3XDg9UyAQ3rfX+utfbJo507HmlkLnNI50JADmwG6S1o0ICRtwAG1kI0bWxt1QzSgcc00Zrddmv1ubIDAFTbVEgAaZxQBGJkWVyr7j0AkMK1GoUvANOYLmMEBYy0BhKjbeOqkWV4vUZo9ByUNAkXrUcHSPw8W6WcHzGSSB6GgnzMIyyKc7KsnTAwQ4VFcnAPkWQRAnoN19pkmReAEVIAACoeAfG3eQe9W6+qJJfe+kqaAf1rEff+zxtaAC8fLTJMEUrgIG3RICUFlXwVKXRi0Rm0JMkgM5xGTEeGgM9S5c7517ldD6qwdZ+nhLQJAfN8AkjdXzCUiczXEccpASD0H6KRlOBkKIzCWN8wkPRSm+JVJAZA/2p9L7+GqmGpxt2YSeCbtA4OVAnrw10FZMzVayGvQ5nQ+rLDhk6q4ftIy0SU7fizthvOy0/Qb77syCRU9+x5EUGsbY5gIslosBWDQSAHcFDweQHRjklB2BOwynwCBcdwuiEgvE7E51eb9Kgeqw5rkvIQLqVlyROkTnkyGrwkaaBuA1nkGvFQE5LbOfBIWIjTAKB0aok7AA3P7Km8XUBB3YKyUI+n4sjQtkOlzhZmDYmrJa6Glmgk/MiQ/TpGRZgLAeKpK4Pn1XhO4OdM4m1ZbxNUgAeXNMlAcBXLUIZsvmzjHcwBHEYN02JGs8CGXfsgCy6XIElZmmpQrxykuTWWf9rpoCcu9Nhhlv7qleHkDoKKOQYlTY/z3Y1i6+A0q73MAfI+W44lBQvqIK+r2fp3yxg/UUT9w0UbfgtellmRIhNlfNeQqkOVeShqs7uv0wTNKCupWIiBkDTuYGy+GoN4WzERZhbCuF8JrsGErgl9kLJ1a9EcaQbRMY0EcgkSVe54T8oJxrYuaBS4UBss2qzoo2CrUUBIhrBY3NnDPQkmSURAoUC4OomwilvM7WYFwM7ykBy/S20H/6Xj+EOvsoL2Ys3Yy2arHO7RGBJejWl+hJFKL5cYsVzilXtExpM6ZfYTE4h0r29gI71SS6Bwrrckr6T27EC7udweiysqvBcDBPm7lvhWXctZALOD+53S4Bj6GcI8eDmJ5F8nmdqf7Pp8zxzu4CKMLItReizFuLle/FV7vFigRMC9yrJAJ0XhWJMidhFCgFyrlpkyXQB55QACczy3kGE+eIFEP4nk/y3IgKHID8XA4KRAkK0KsKYARgo4WQeeaKCuWKSubQCBbmpSEM/KZE8yHUsgZEky+KtEDExKB87EJu5KwENo1K/EtyH8tu9AfIP8iknsCBo0GBcIS6pEnOmBMyqYmBXIYSJEDqwoGAJEkY8Q9kikV4o06iYukwRARAQouydAgQkE8eNGCUhYqkKuDA4MXK8Q40wOJA9kk6Ch3KGgCqj6y6wSFkAA3vRMpiCq3h4PRL7irCQAAL7HqUY/Q2Z4pFJTJYHlKVILJdRcG8EDAoj64WS+HlYi5hItwPCUE/KJLtYkBgAUJCpBBZ6VSmG/gxH0Cra1RGwo4PCiGRCSFZqhpeqZHOG4BgC9q/pq4xg1hEYPB8RrxgjYBiDmbHRhhiHRDEgeDZI26dihyMyY5pSqarSIAnpYFkpk6/IkD0x8CYzdznQIHIBkhCiayKDkZ0A477ysT47Xw/RE6myXybrKS3whj3wgrU5xbPx06ZCfaWYeLxIIFgCWgxZ4R0a3HW5P7kAv70G0APKqjFDPJSgfJfIAGihAFCggFARArgGQCBB0DwCOBEowr2hITbFIF76F54roFVgkS74K4EqkEwokoUGbhUGUpgR8S0qkD0qBLFQ/LsFLYgorYvClEbZ+6KSKZORBHMCR50Jq6qRRCyqDoy7b5OZ7HcnwiRTqCEifCsiCkZ5dZSkymMA57aAaA2JB7spEEoDMCfAgqQCinil2Jw427M70A/yzZhJxLAqjQWFGHWF9S2EihnKGn5GGEzSXqQAABCgxVRIxqYLSmsFWvKWmXgbQdGtI2mIqYqTqUYXAqkDRrhHh6akA0Iv09EEA8Z9RWMYSTRam/a9EgMak2ZzR6mXADoA460BZRZWRhAJZHqZZIKFZD6bhgMeGeS3C/w4ZWApCbAVI6UBkMkUZzgDw/WFEooma9a6aYAmI56QYkhOIqkeut6FArhsm3gkAAALPmZAIWRALuZ6L2b+rnM+iQPRKyPRMedWaXp/KJGjp3pkCCG7vsGJJfhpOjCjtQONirD6DzGLnZmgmvrplQNwCgA3FgDzJqeqgkeRPCDubAk5nmI1q7hmGqvwHgAMN4MMlAISCIBLLmAZEQHVGTjsYzDrDep6MsJ+odCQImdQGgDeepv8gAI7YDwBChG4YzOj3BLmpoOwpkdxfajz+FNkeBWrdYSb8jVBjCTAkSjBXAjAhl+DvIkRjDvKjD8jvJXCzSyoEy3wYAxaKjBYxQJCcY7lMX7n1kybgaQYShZZ7Z4hsUHluUeVAniLkUnhup8YIAYCCb7nCZrASDiZYADBqBHT+xIZDawBeycW+X3nCmjR1mVkyRq6yVuryVWoBV5JBWUVJlxDKEdQ/I96MUpDXkuUyS2wRQtwgW0CWiACYBPCI3skklfpt0djMqTFKFpkPEgVaNEVQUW6ZXNDh4JBoqFlZJiYe5YhvpqedyIklJXZfwEhbbBkYPEoCVUmAwPcOaJLFhO2X8hVUoSoTVeEA8EjHai6gHKNNaT5j7NiEwvuRQmZROFrsgA4ByDjOwLZdUGOKWrVegKaPgIvhYmNfXM3HIZ4fZG6RKAddIKyMKK8BwU5RlRFJBseatalapA4l9c4iQATXacCdcq/uCZAPcp/s8qqLCf/j8oiQCiiWASClwBiXRtiTAXiQhI+MhpxnxO+IQKkYAbuP+B9GgEBIyfIMjp5NeGoDBPePBAYMLdEeoNNCREicSGSBDKkM4PCHOJrQuJAKqLQFEGUMeYUAwMeR/rKOUOUIJkRqqLKAILQC7QIMUAIIqMKOUIqB/lEOULKFKBUFKHBELRbdrbgLrfraSHQCREuNHebRANTF5BOaQHhWeowsbZaWbfYcMvREgLYCGXMMRrQNMkblYDDduO4dEM2asSXUgCdlIJrl/BgI3VEM3cyCXZ5DYCoKrQAOqyo0BWAUDuC4BeA9190l2nDzCjEV34Bnq2Bz0KUt1VD0R0a0D+4YBjCr1tSBiID8i/kb07D93b27373T1eBn0ZiN1pDppX2QA71HD73GiICmgfDKQP1npP2eGv30SHT6q0AjAi7pqIBtSN2MTAPC64D/37CEgOBHyICN2/TDJVDF1VC4Nv3ua5BoBsCwN32Wzn2v14P0TG1fToNcDP1b14Nv1uy1hk4kOmziAz0PAADkcYwQM+jqW5os6OBYiAXDepwoFAMWWubx05yRMZMYLJlCwxzo50lREYzq8sjkh0QY5WPYuqwjwio6TeNsMYvDGgj5WDuD9E1hsDJIzgasVEFjjDb93olo1F5mF9DDlDnoSSGQA4SDhDxDXA9EHDs9ljXhFDkAODjD9EBDRDD5wTX9P952Gs7mTjMT1D2IgDL9lj29zDmArDwTt0k839Zo70dG3oM98g/xNMQFa8ijSMmNWMsQWiuqa8A8KIMYggFFp4SjBkKj9qo56ZQYAktA2jos/x50Bjo17m0YWAZj6TlDNjwTdjY8gYizVjrj4VSSHjXAvdm9kT29PjGG/jv5gTCTb9RsyTykMKeDXhwyxCJdCDtgSTZTyUsDaoJAx5x5xQH+XzDtH+aAxQsopQ1ttAsoH+rG1tUQDAqoUomox5JAsoiLDAiocotAIdQdJAqoPzsoaAaAcoH+H+pQAdxQdt5QpQ6TID12tgpDsDDA1try2LpQqoqLx5HtZLUQUoAgEd9wFQvdx5n+KgDAEox54LpQaAodUoDtUQeLoLqo5QSgAgH+QdaoiopQUQ6rMK9z6dUAekJE2dXkszKdnGD4Ft4tG6WcuBOkBdptGtBgjhzza4OwRiuAhIyJdANd6gjU8G7hUoOrwtFrqMVr1DXkqd+gQAA=== -->\n\n<!-- internal state end -->"},"request":{"signal":{}}},"response":{"url":"https://api.github.com/repos/NVIDIA/cloudai/issues/comments/4163563183","status":401,"headers":{"access-control-allow-origin":"*","access-control-expose-headers":"ETag, Link, Location, Retry-After, X-GitHub-OTP, X-RateLimit-Limit, X-RateLimit-Remaining, X-RateLimit-Used, X-RateLimit-Resource, X-RateLimit-Reset, X-OAuth-Scopes, X-Accepted-OAuth-Scopes, X-Poll-Interval, X-GitHub-Media-Type, X-GitHub-SSO, X-GitHub-Request-Id, Deprecation, Sunset","connection":"close","content-security-policy":"default-src 'none'","content-type":"application/json; charset=utf-8","date":"Tue, 31 Mar 2026 15:41:14 GMT","referrer-policy":"origin-when-cross-origin, strict-origin-when-cross-origin","server":"github.com","strict-transport-security":"max-age=31536000; includeSubdomains; preload","vary":"Accept-Encoding, Accept, X-Requested-With","x-content-type-options":"nosniff","x-frame-options":"deny","x-github-media-type":"github.v3; format=json","x-github-request-id":"4693:C7704:2CBB27:B6E556:69CBEB19","x-xss-protection":"0"},"data":{"message":"Requires authentication","documentation_url":"https://docs.github.com/rest","status":"401"}}}

[coderabbitai]

Actionable comments posted: 1

🤖 Prompt for all review comments with AI agents

Verify each finding against the current code and only fix it if needed.

Inline comments:
In `@src/cloudai/workloads/vllm/vllm.py`:
- Around line 96-129: The code in _validate_vllm_parallelism_constraints uses a
nested getattr to fetch expert_parallel with a fallback to
enable_expert_parallel, which adds unnecessary complexity; change the ep_enabled
assignment to directly read the canonical field (e.g., ep_enabled = cast(bool,
getattr(args, "enable_expert_parallel", False))) and remove the nested getattr;
if expert_parallel is intended as a supported alias instead, add documentation
and update tests to reflect that alias rather than keeping the nested lookup in
validate logic.

🪄 Autofix (Beta)

Fix all unresolved CodeRabbit comments on this PR:

• Push a commit to this branch (recommended)
• Create a new PR with the fixes

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: ASSERTIVE

Plan: Pro

Run ID: 517f1db1-3c09-4041-80af-b82be6b2c21e

📥 Commits

Reviewing files that changed from the base of the PR and between 7d0416f and 609a025.

📒 Files selected for processing (3)

• src/cloudai/workloads/common/llm_serving.py
• src/cloudai/workloads/vllm/vllm.py
• tests/workloads/vllm/test_workload.py

[coderabbitai]

Actionable comments posted: 1

🤖 Prompt for all review comments with AI agents

Verify each finding against the current code and only fix it if needed.

Inline comments:
In `@src/cloudai/workloads/vllm/vllm.py`:
- Around line 98-104: The code uses cast(...) on args fields (tp, pp, dp) but if
the attribute exists and is None the arithmetic at constraint1 = (tp * pp * dp)
<= gpu_count will raise TypeError; update the logic in the vllm workload setup
to normalize/validate these parallelism values from args before using them
(e.g., check args.tensor_parallel_size, args.pipeline_parallel_size,
args.data_parallel_size for None and replace with the intended default or
raise/return an error), and ensure ep_enabled and all2all_backend handling
remains unchanged; concretely, validate/convert tp, pp, dp to ints (or bail with
a clear error) prior to computing constraint1 so multiplication never receives
None.

🪄 Autofix (Beta)

Fix all unresolved CodeRabbit comments on this PR:

• Push a commit to this branch (recommended)
• Create a new PR with the fixes

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: ASSERTIVE

Plan: Pro

Run ID: 8ab3f09f-d909-4fa1-8646-c7859874fbb9

📥 Commits

Reviewing files that changed from the base of the PR and between 609a025 and c5d2bac.

📒 Files selected for processing (1)

• src/cloudai/workloads/vllm/vllm.py