{"id":"GHSA-9g3c-x747-c6wr","kind":"vulnerability","aliases":["GHSA-9g3c-x747-c6wr"],"related":[],"title":"Quadratic Cohere2 Vision preprocessing DoS via request-controlled max_patches","description":"## Summary\n\nThe Python OpenAI-compatible server accepts a request-controlled\n`mm_processor_kwargs.max_patches` value when serving Cohere2 Vision models.\nvLLM forwards the value to the Hugging Face processor without applying a\nserver-side upper bound. The processor enumerates candidate tile aspect ratios\nwith two loops whose bounds are both controlled by `max_patches`, causing\nquadratic CPU work before normal inference limits can protect the service.\n\nA client that can access the chat-completions endpoint can submit a small,\nvalid image together with a large `max_patches` value. Varying the value across\nrequests also avoids reuse of the helper's small LRU cache. This can occupy\nmultimodal preprocessing workers, substantially increase request latency, and\ndeny service to other clients.\n\nNo code execution, confidentiality impact, or integrity impact was\ndemonstrated.\n\n## Confirmed Affected Revision\n\nThe issue was rechecked on 2026-09-08 against the then-current upstream\n`main` commit:\n\n```text\nf6326f53bda46898a331c2d24500332c285d9a2b\n```\n\nCommit permalink:\nhttps://github.com/vllm-project/vllm/commit/f6326f53bda46898a331c2d24500332c285d9a2b\n\nThe original audited revision was:\n\n```text\n7ca49fbe4bab019e55d57cdc4b7fd3d55c67c1a6\n```\n\nThe vulnerable data flow was present at the revision assessed in this report.\n\n## Why This Is a Security Issue\n\n`max_patches` is a computational budget, but the Python serving path lets a\nremote request choose it. The expensive enumeration occurs during request\npreprocessing, before the request reaches normal model execution. A small\nrequest body can therefore trigger work disproportionate to its size.\n\nThe default server does not require an API key. If an operator configures API\nauthentication, exploitation instead requires a client authorized to call the\naffected endpoint.\n\n## Pre-requisites\n\n- vLLM is serving a Cohere2 Vision model through the Python OpenAI-compatible\n  frontend.\n- The caller can submit an image chat request.\n- The request supplies a large positive integer as\n  `mm_processor_kwargs.max_patches`.\n\nNo special build flag, remote model code, malformed image, or unsafe runtime\noption is required. The issue is model-specific and is not a claim that every\nvLLM deployment is affected.\n\n## Technical Details\n\nThe request-to-sink chain at the confirmed revision is:\n\n1. `ChatCompletionRequest.mm_processor_kwargs` is an unrestricted\n   `dict[str, Any]`:\n   https://github.com/vllm-project/vllm/blob/f6326f53bda46898a331c2d24500332c285d9a2b/vllm/entrypoints/openai/chat_completion/protocol.py#L372-L375\n2. `OnlineRenderer.preprocess_chat` copies the request value into both the chat\n   parameters and prompt extras:\n   https://github.com/vllm-project/vllm/blob/f6326f53bda46898a331c2d24500332c285d9a2b/vllm/renderers/online_renderer.py#L675-L704\n3. `BaseRenderer` applies the extras and constructs `MMProcessorInputs` with\n   the same processor kwargs:\n   https://github.com/vllm-project/vllm/blob/f6326f53bda46898a331c2d24500332c285d9a2b/vllm/renderers/base.py#L769-L779\n   https://github.com/vllm-project/vllm/blob/f6326f53bda46898a331c2d24500332c285d9a2b/vllm/renderers/base.py#L846-L880\n4. `Cohere2VisionProcessingInfo.get_num_patches` passes the merged mapping to\n   `get_number_of_image_patches` without bounding `max_patches`:\n   https://github.com/vllm-project/vllm/blob/f6326f53bda46898a331c2d24500332c285d9a2b/vllm/model_executor/models/cohere2_vision.py#L170-L188\n5. In Transformers 5.16.1, `get_number_of_image_patches` reads\n   `max_patches` from that mapping and calls `get_optimal_tiled_canvas`:\n   https://github.com/huggingface/transformers/blob/v5.16.1/src/transformers/models/cohere2_vision/image_processing_cohere2_vision.py#L257-L285\n6. The aspect-ratio helper loops from 1 through the attacker-selected value in\n   both dimensions:\n   https://github.com/huggingface/transformers/blob/v5.16.1/src/transformers/models/cohere2_vision/image_processing_cohere2_vision.py#L53-L80\n\nThe relevant operation is equivalent to:\n\n```python\nfor width in range(1, max_image_tiles + 1):\n    for height in range(1, max_image_tiles + 1):\n        if width * height <= max_image_tiles:\n            aspect_ratios.append((width, height))\n```\n\nThe nested loops perform Theta(`max_patches` squared) iterations. The helper\nhas an LRU cache of only ten argument values, so requests using distinct large\nvalues continue to force cold computations.\n\n## Expected Result\n\nRequest-level processor overrides must not be able to raise a server-owned\npreprocessing budget beyond a safe configured limit. An excessive value should\nbe rejected before invoking the Hugging Face processor.\n\n## Security Impact\n\nA remote or API-authorized caller can consume disproportionate CPU time in the\nmultimodal preprocessing path. Repeated cold computations can delay image\nrequests and reduce or deny availability for other clients. The demonstrated\nimpact is limited to availability.\n\n## Severity Assessment\n\n- Severity: Moderate\n- CWE-407: Inefficient Algorithmic Complexity\n- CWE-400: Uncontrolled Resource Consumption\n- CVSS v3.1: 5.3\n- Vector: `CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:N/I:N/A:L`\n\n`PR:N` reflects the default server configuration without an API key. An\nauthenticated deployment changes the required access conditions, and the vLLM\nteam should assign the final severity and version range.\n\n## Duplicate Search\n\nOn 2026-09-08, public vLLM issues, pull requests, and published GitHub Security\nAdvisories were searched using combinations of `max_patches`, `Cohere2\nVision`, `get_all_supported_aspect_ratios`, `mm_processor_kwargs`, `quadratic`,\nand `denial of service`. No public report with this specific model, request\nparameter, and root cause was found.\n\nPrivate vulnerability reports are not searchable, so this does not rule out a\nprivate duplicate.\n\n## Suggested Remediation\n\n- Treat the server/model configuration as the authority for the maximum\n  permitted patch count.\n- Reject request-level `max_patches` values that are not strict positive\n  integers or that exceed that server-owned limit.\n- Apply the validation to both flat and modality-scoped processor kwargs where\n  supported.\n- Add tests proving that an oversized value is rejected before Cohere2 Vision\n  preprocessing and that safe values continue to work.\n- Consider improving the upstream Transformers enumeration independently, but\n  do not rely on that optimization as the request-boundary security control.\n\n## Disclosure Coordination\n\nPlease handle this report through vLLM's private vulnerability-reporting\nworkflow. No public issue or pull request has been created for this finding.\nThe reporter can provide additional source traces or help validate a proposed\nfix. Public disclosure and any public remediation pull request should follow\nthe maintainers' triage and coordination guidance.\n\n## Affected versions\n\n- Package: `vllm` (PyPI).\n- Affected versions: `>= 0.10.1, < 0.31.0`.\n- First stable patched version: `0.31.0`.\n\nPR #58830 blocks untrusted per-request `mm_processor_kwargs` and `media_io_kwargs` by default before preprocessing. Keep `--trust-request-mm-kwargs` disabled for untrusted clients. Explicitly enabling it restores trusted override behavior; the underlying processor/sampler is not independently bounded by this fix.\n\nAdvisory fixed in [#58830](https://github.com/vllm-project/vllm/pull/58830).","summary":"A client sending an image chat request to vLLM serving a Cohere2 Vision model can set mm_processor_kwargs.max_patches to a large value, forcing quadratic CPU work during preprocessing. This lets an attacker slow or deny service to other clients, with no code execution or data exposure demonstrated.","severity":"medium","cvss":null,"epss":null,"exploited":false,"exploited_at":null,"products":[{"name":"vllm","fixed":["0.31.0"],"affected":">= 0.10.1, < 0.31.0","ecosystem":"pypi"}],"ecosystems":["pypi"],"tags":["ai-tooling","cwe-400","cwe-407","llm-serving","python","self-hosted","vllm"],"not_affected_if":["api-auth-with-trusted-clients","no-cohere2-vision-model","version-0.31.0-or-later"],"how_to_check":"- Check the installed vLLM version with `pip show vllm` or `python -c \"import vllm; print(vllm.__version__)\"`; versions from `>= 0.10.1` and below `0.31.0` are affected.\n- Check whether the server is serving a Cohere2 Vision model through the OpenAI-compatible server.\n- Check whether the server accepts per-request `mm_processor_kwargs` from untrusted clients; the advisory says `--trust-request-mm-kwargs` should stay disabled for them.","fix":"- Upgrade `vllm` to `0.31.0` or later.\n- Keep `--trust-request-mm-kwargs` disabled for untrusted clients (per PR #58830, untrusted per-request `mm_processor_kwargs` are blocked by default).","status":"open","published_at":"2026-10-27T22:20:43+00:00","first_seen_at":"2026-11-09T22:53:23+00:00","triaged_at":"2026-11-09T22:58:55+00:00","updated_at":"2026-11-09T22:58:55+00:00","category":"ai-tooling","confidence":0.8,"triage":{"at":"2026-11-09T22:58:55Z","model":"claude-haiku-5-5","output":{"fix":"- Upgrade `vllm` to `0.31.0` or later.\n- Keep `--trust-request-mm-kwargs` disabled for untrusted clients (per PR #58830, untrusted per-request `mm_processor_kwargs` are blocked by default).","reach":"internet","title":"vLLM: Cohere2 Vision image requests can exhaust CPU via max_patches","impact":"low","summary":"A client sending an image chat request to vLLM serving a Cohere2 Vision model can set mm_processor_kwargs.max_patches to a large value, forcing quadratic CPU work during preprocessing. This lets an attacker slow or deny service to other clients, with no code execution or data exposure demonstrated.","audience":"businesses","category":"ai-tooling","nicknames":[],"who_cares":["vllm","python","self-hosted","llm-serving"],"confidence":0.8,"fix_status":"available","needs_auth":"no","worst_case":"Remote clients exhaust CPU in multimodal preprocessing, denying service to others","analyst_note":"","how_to_check":"- Check the installed vLLM version with `pip show vllm` or `python -c \"import vllm; print(vllm.__version__)\"`; versions from `>= 0.10.1` and below `0.31.0` are affected.\n- Check whether the server is serving a Cohere2 Vision model through the OpenAI-compatible server.\n- Check whether the server accepts per-request `mm_processor_kwargs` from untrusted clients; the advisory says `--trust-request-mm-kwargs` should stay disabled for them.","plain_action":"","product_name":"","needs_analyst":false,"not_affected_if":["no-cohere2-vision-model","api-auth-with-trusted-clients","version-0.31.0-or-later"],"who_is_affected":"Teams running vLLM's Python OpenAI-compatible server with a Cohere2 Vision model, where clients can send image chat requests. The default server has no API key, so any reachable client can trigger it; deployments with API authentication limit this to authorized clients. Deployments not serving Cohere2 Vision models are not affected.","user_interaction":"no","product_ecosystem":"","needs_special_setup":"yes","severity_if_unknown":"medium"},"cost_usd":0.001113,"input_tokens":7316,"output_tokens":763,"severity_before":"medium"},"analyst_status":null,"analyst_note":null,"search":"'-08':226C '-09':225C '-400':27B '-407':29B '/huggingface/transformers/blob/v5.16.1/src/transformers/models/cohere2_vision/image_processing_cohere2_vision.py#l257-l285':511C '/huggingface/transformers/blob/v5.16.1/src/transformers/models/cohere2_vision/image_processing_cohere2_vision.py#l53-l80':532C '/vllm-project/vllm/blob/f6326f53bda46898a331c2d24500332c285d9a2b/vllm/entrypoints/openai/chat_completion/protocol.py#l372-l375':429C '/vllm-project/vllm/blob/f6326f53bda46898a331c2d24500332c285d9a2b/vllm/model_executor/models/cohere2_vision.py#l170-l188':487C '/vllm-project/vllm/blob/f6326f53bda46898a331c2d24500332c285d9a2b/vllm/renderers/base.py#l769-l779':463C '/vllm-project/vllm/blob/f6326f53bda46898a331c2d24500332c285d9a2b/vllm/renderers/base.py#l846-l880':466C '/vllm-project/vllm/blob/f6326f53bda46898a331c2d24500332c285d9a2b/vllm/renderers/online_renderer.py#l675-l704':447C '/vllm-project/vllm/commit/f6326f53bda46898a331c2d24500332c285d9a2b':241C '1':417C,520C '2':430C '2026':224C '3':448C '4':467C '5':488C '5.16.1':491C '6':512C '7ca49fbe4bab019e55d57cdc4b7fd3d55c67c1a6':248C '9g3c':14A,19A 'accept':92C 'access':156C 'across':178C 'affect':217C,335C,403C 'ai':24B 'ai-tool':23B 'also':180C 'api':318C,324C 'appli':115C,450C 'aspect':127C,515C 'aspect-ratio':514C 'assess':258C 'attack':69B,524C 'attacker-select':523C 'audit':244C 'authent':325C 'author':331C 'avoid':181C 'baserender':449C 'bodi':302C 'bound':121C,133C,482C 'budget':273C 'build':374C 'c6wr':16A,21A 'cach':189C 'call':333C,504C 'caller':355C 'candid':125C 'canva':508C 'caus':140C 'chain':411C 'chat':43B,159C,360C,432C,440C 'chat-complet':158C 'chatcompletionrequest.mm':418C 'choos':283C 'claim':397C 'client':39B,76B,153C,205C,330C 'code':79B,207C,378C 'cohere2':2A,49B,102C,344C 'cohere2visionprocessinginfo.get':468C 'commit':234C,237C 'compat':90C,352C 'complet':160C 'comput':272C 'confidenti':209C 'configur':323C 'confirm':216C,414C 'construct':454C 'control':9A,96C,136C 'copi':433C 'cpu':62B,142C 'current':231C 'cwe':26B,28B 'data':82B,251C 'default':312C 'demonstr':84B,215C 'deni':72B,201C 'deploy':401C 'detail':405C 'dict':424C 'dimens':529C 'disproportion':307C 'dos':5A 'endpoint':161C,336C 'enumer':124C,287C 'equiv':537C 'everi':399C 'execut':80B,208C,298C 'expens':286C 'exploit':326C 'exposur':83B 'extra':444C,452C 'f6326f53bda46898a331c2d24500332c285d9a2b':236C 'face':112C 'flag':375C 'flow':252C 'forc':60B 'forward':106C 'frontend':353C 'get':476C,492C,505C 'ghsa':13A,18A 'ghsa-9g3c-x747-c6wr':12A,17A 'github.com':240C,428C,446C,462C,465C,486C,510C,531C 'github.com/huggingface/transformers/blob/v5.16.1/src/transformers/models/cohere2_vision/image_processing_cohere2_vision.py#l257-l285':509C 'github.com/huggingface/transformers/blob/v5.16.1/src/transformers/models/cohere2_vision/image_processing_cohere2_vision.py#l53-l80':530C 'github.com/vllm-project/vllm/blob/f6326f53bda46898a331c2d24500332c285d9a2b/vllm/entrypoints/openai/chat_completion/protocol.py#l372-l375':427C 'github.com/vllm-project/vllm/blob/f6326f53bda46898a331c2d24500332c285d9a2b/vllm/model_executor/models/cohere2_vision.py#l170-l188':485C 'github.com/vllm-project/vllm/blob/f6326f53bda46898a331c2d24500332c285d9a2b/vllm/renderers/base.py#l769-l779':461C 'github.com/vllm-project/vllm/blob/f6326f53bda46898a331c2d24500332c285d9a2b/vllm/renderers/base.py#l846-l880':464C 'github.com/vllm-project/vllm/blob/f6326f53bda46898a331c2d24500332c285d9a2b/vllm/renderers/online_renderer.py#l675-l704':445C 'github.com/vllm-project/vllm/commit/f6326f53bda46898a331c2d24500332c285d9a2b':239C 'helper':185C,517C 'host':36B 'hug':111C 'imag':42B,167C,359C,380C,479C,495C 'impact':210C,213C 'increas':197C 'infer':146C 'instead':327C 'integ':368C 'integr':212C 'issu':220C,267C,388C 'key':319C 'kwarg':420C,460C 'larg':58B,171C,366C 'latenc':199C 'let':67B,279C 'limit':147C 'llm':31B 'llm-serv':30B 'loop':131C,518C 'lru':188C 'main':233C 'malform':379C 'map':474C,502C 'max':10A,138C,172C,268C,483C,498C 'merg':473C 'mm_processor_kwargs.max':54B,97C,370C 'mmprocessorinput':455C 'model':51B,104C,297C,346C,377C,391C 'model-specif':390C 'multimod':193C 'normal':145C,296C 'num':469C 'number':477C,493C 'occupi':192C 'occur':288C 'onlinerenderer.preprocess':431C 'openai':89C,351C 'openai-compat':88C,350C 'oper':322C,535C 'optim':506C 'option':384C 'origin':243C 'paramet':441C 'pass':471C 'patch':11A,55B,98C,139C,173C,269C,371C,470C,480C,484C,496C,499C 'path':278C 'permalink':238C 'posit':367C 'pre':338C 'pre-requisit':337C 'preprocess':4A,65B,194C,291C 'present':254C 'processor':113C,123C,419C,459C 'prompt':443C 'protect':149C 'python':33B,87C,276C,349C 'quadrat':1A,61B,141C 'ratio':128C,516C 'reach':295C 'read':497C 'recheck':222C 'relev':534C 'remot':281C,376C 'report':261C 'request':8A,44B,95C,179C,198C,282C,290C,294C,301C,361C,363C,408C,435C 'request-control':7A,94C 'request-to-sink':407C 'requir':316C,328C,386C 'requisit':339C 'reus':182C 'revis':218C,245C,257C,415C 'runtim':383C 'secur':266C 'select':525C 'self':35B 'self-host':34B 'send':40B 'serv':32B,47B,101C,277C,342C 'server':91C,118C,313C 'server-sid':117C 'servic':73B,151C,202C 'set':53B 'side':119C 'sink':410C 'size':310C 'slow':70B 'small':165C,187C,300C 'special':373C 'specif':392C 'str':425C 'submit':163C,357C 'substanti':196C 'summari':85C 'suppli':364C 'technic':404C 'text':235C,247C 'then-curr':229C 'therefor':304C 'tile':126C,507C 'togeth':168C 'tool':25B 'transform':490C 'trigger':305C 'two':130C 'unrestrict':423C 'unsaf':382C 'upper':120C 'upstream':232C 'valid':166C 'valu':59B,99C,108C,174C,177C,436C,526C 'vari':175C 'via':6A 'vision':3A,50B,103C,345C 'vllm':22A,37B,46B,105C,340C,400C 'vulner':250C 'whose':132C 'without':114C,481C 'work':63B,143C,306C 'worker':195C 'x747':15A,20A","nicknames":[],"mitigations":[],"who_affected":"Teams running vLLM's Python OpenAI-compatible server with a Cohere2 Vision model, where clients can send image chat requests. The default server has no API key, so any reachable client can trigger it; deployments with API authentication limit this to authorized clients. Deployments not serving Cohere2 Vision models are not affected.","tier":"B","verified":{},"analyst_revisit_at":null,"fix_status":"available","audience":"businesses","plain_action":null,"pkeys":["vllm"],"ptokens":["vllm"],"labels":["aitooling","cwe400","cwe407","llmserving","medium","pypi","python","selfhosted","vllm"],"campaign_type":null,"search_a":"'9g3c':14A,19A 'c6wr':16A,21A 'cohere2':2A 'control':9A 'dos':5A 'ghsa':13A,18A 'ghsa-9g3c-x747-c6wr':12A,17A 'max':10A 'patch':11A 'preprocess':4A 'quadrat':1A 'request':8A 'request-control':7A 'via':6A 'vision':3A 'vllm':22A 'x747':15A,20A","search_b":"'-400':27B '-407':29B 'ai':24B 'ai-tool':23B 'attack':69B 'chat':43B 'client':39B,76B 'code':79B 'cohere2':49B 'cpu':62B 'cwe':26B,28B 'data':82B 'demonstr':84B 'deni':72B 'execut':80B 'exposur':83B 'forc':60B 'host':36B 'imag':42B 'larg':58B 'let':67B 'llm':31B 'llm-serv':30B 'mm_processor_kwargs.max':54B 'model':51B 'patch':55B 'preprocess':65B 'python':33B 'quadrat':61B 'request':44B 'self':35B 'self-host':34B 'send':40B 'serv':32B,47B 'servic':73B 'set':53B 'slow':70B 'tool':25B 'valu':59B 'vision':50B 'vllm':37B,46B 'work':63B","search_c":"'-08':226C '-09':225C '/huggingface/transformers/blob/v5.16.1/src/transformers/models/cohere2_vision/image_processing_cohere2_vision.py#l257-l285':511C '/huggingface/transformers/blob/v5.16.1/src/transformers/models/cohere2_vision/image_processing_cohere2_vision.py#l53-l80':532C '/vllm-project/vllm/blob/f6326f53bda46898a331c2d24500332c285d9a2b/vllm/entrypoints/openai/chat_completion/protocol.py#l372-l375':429C '/vllm-project/vllm/blob/f6326f53bda46898a331c2d24500332c285d9a2b/vllm/model_executor/models/cohere2_vision.py#l170-l188':487C '/vllm-project/vllm/blob/f6326f53bda46898a331c2d24500332c285d9a2b/vllm/renderers/base.py#l769-l779':463C '/vllm-project/vllm/blob/f6326f53bda46898a331c2d24500332c285d9a2b/vllm/renderers/base.py#l846-l880':466C '/vllm-project/vllm/blob/f6326f53bda46898a331c2d24500332c285d9a2b/vllm/renderers/online_renderer.py#l675-l704':447C '/vllm-project/vllm/commit/f6326f53bda46898a331c2d24500332c285d9a2b':241C '1':417C,520C '2':430C '2026':224C '3':448C '4':467C '5':488C '5.16.1':491C '6':512C '7ca49fbe4bab019e55d57cdc4b7fd3d55c67c1a6':248C 'accept':92C 'access':156C 'across':178C 'affect':217C,335C,403C 'also':180C 'api':318C,324C 'appli':115C,450C 'aspect':127C,515C 'aspect-ratio':514C 'assess':258C 'attack':524C 'attacker-select':523C 'audit':244C 'authent':325C 'author':331C 'avoid':181C 'baserender':449C 'bodi':302C 'bound':121C,133C,482C 'budget':273C 'build':374C 'cach':189C 'call':333C,504C 'caller':355C 'candid':125C 'canva':508C 'caus':140C 'chain':411C 'chat':159C,360C,432C,440C 'chat-complet':158C 'chatcompletionrequest.mm':418C 'choos':283C 'claim':397C 'client':153C,205C,330C 'code':207C,378C 'cohere2':102C,344C 'cohere2visionprocessinginfo.get':468C 'commit':234C,237C 'compat':90C,352C 'complet':160C 'comput':272C 'confidenti':209C 'configur':323C 'confirm':216C,414C 'construct':454C 'control':96C,136C 'copi':433C 'cpu':142C 'current':231C 'data':251C 'default':312C 'demonstr':215C 'deni':201C 'deploy':401C 'detail':405C 'dict':424C 'dimens':529C 'disproportion':307C 'endpoint':161C,336C 'enumer':124C,287C 'equiv':537C 'everi':399C 'execut':208C,298C 'expens':286C 'exploit':326C 'extra':444C,452C 'f6326f53bda46898a331c2d24500332c285d9a2b':236C 'face':112C 'flag':375C 'flow':252C 'forward':106C 'frontend':353C 'get':476C,492C,505C 'github.com':240C,428C,446C,462C,465C,486C,510C,531C 'github.com/huggingface/transformers/blob/v5.16.1/src/transformers/models/cohere2_vision/image_processing_cohere2_vision.py#l257-l285':509C 'github.com/huggingface/transformers/blob/v5.16.1/src/transformers/models/cohere2_vision/image_processing_cohere2_vision.py#l53-l80':530C 'github.com/vllm-project/vllm/blob/f6326f53bda46898a331c2d24500332c285d9a2b/vllm/entrypoints/openai/chat_completion/protocol.py#l372-l375':427C 'github.com/vllm-project/vllm/blob/f6326f53bda46898a331c2d24500332c285d9a2b/vllm/model_executor/models/cohere2_vision.py#l170-l188':485C 'github.com/vllm-project/vllm/blob/f6326f53bda46898a331c2d24500332c285d9a2b/vllm/renderers/base.py#l769-l779':461C 'github.com/vllm-project/vllm/blob/f6326f53bda46898a331c2d24500332c285d9a2b/vllm/renderers/base.py#l846-l880':464C 'github.com/vllm-project/vllm/blob/f6326f53bda46898a331c2d24500332c285d9a2b/vllm/renderers/online_renderer.py#l675-l704':445C 'github.com/vllm-project/vllm/commit/f6326f53bda46898a331c2d24500332c285d9a2b':239C 'helper':185C,517C 'hug':111C 'imag':167C,359C,380C,479C,495C 'impact':210C,213C 'increas':197C 'infer':146C 'instead':327C 'integ':368C 'integr':212C 'issu':220C,267C,388C 'key':319C 'kwarg':420C,460C 'larg':171C,366C 'latenc':199C 'let':279C 'limit':147C 'loop':131C,518C 'lru':188C 'main':233C 'malform':379C 'map':474C,502C 'max':138C,172C,268C,483C,498C 'merg':473C 'mm_processor_kwargs.max':97C,370C 'mmprocessorinput':455C 'model':104C,297C,346C,377C,391C 'model-specif':390C 'multimod':193C 'normal':145C,296C 'num':469C 'number':477C,493C 'occupi':192C 'occur':288C 'onlinerenderer.preprocess':431C 'openai':89C,351C 'openai-compat':88C,350C 'oper':322C,535C 'optim':506C 'option':384C 'origin':243C 'paramet':441C 'pass':471C 'patch':98C,139C,173C,269C,371C,470C,480C,484C,496C,499C 'path':278C 'permalink':238C 'posit':367C 'pre':338C 'pre-requisit':337C 'preprocess':194C,291C 'present':254C 'processor':113C,123C,419C,459C 'prompt':443C 'protect':149C 'python':87C,276C,349C 'quadrat':141C 'ratio':128C,516C 'reach':295C 'read':497C 'recheck':222C 'relev':534C 'remot':281C,376C 'report':261C 'request':95C,179C,198C,282C,290C,294C,301C,361C,363C,408C,435C 'request-control':94C 'request-to-sink':407C 'requir':316C,328C,386C 'requisit':339C 'reus':182C 'revis':218C,245C,257C,415C 'runtim':383C 'secur':266C 'select':525C 'serv':101C,277C,342C 'server':91C,118C,313C 'server-sid':117C 'servic':151C,202C 'side':119C 'sink':410C 'size':310C 'small':165C,187C,300C 'special':373C 'specif':392C 'str':425C 'submit':163C,357C 'substanti':196C 'summari':85C 'suppli':364C 'technic':404C 'text':235C,247C 'then-curr':229C 'therefor':304C 'tile':126C,507C 'togeth':168C 'transform':490C 'trigger':305C 'two':130C 'unrestrict':423C 'unsaf':382C 'upper':120C 'upstream':232C 'valid':166C 'valu':99C,108C,174C,177C,436C,526C 'vari':175C 'vision':103C,345C 'vllm':105C,340C,400C 'vulner':250C 'whose':132C 'without':114C,481C 'work':143C,306C 'worker':195C","cause_entry":null,"cause_path":null,"disclosures_checked_at":null,"first_public_at":"2026-10-27T22:20:43+00:00","kev":null,"cvss_vector":null,"cvss_by":null,"poc_count":0,"sources":[{"source":"ghsa-repo","external_id":"GHSA-9g3c-x747-c6wr","url":"https://github.com/vllm-project/vllm/security/advisories/GHSA-9g3c-x747-c6wr","severity":"medium","published_at":"2026-10-27T22:20:43+00:00","modified_at":"2026-10-27T22:20:43+00:00","first_seen_at":"2026-11-09T22:53:23+00:00","fetched_at":"2026-11-09T22:53:23+00:00","note":null}],"documents":[{"id":110585,"url":"https://github.com/vllm-project/vllm/security/advisories/GHSA-9g3c-x747-c6wr","title":"Maintainer's advisory: github.com","kind":"vendor","source":"ghsa-repo","publisher":"vendor","text":null,"published_at":null,"relation":"advisory","confidence":null,"note":"Maintainer's advisory: github.com"}],"disclosures":[],"exploit_sources":[],"notice":"Free to read, with fair use of about 60 pages a minute. For higher limits or API access: sales@infrastack.ai"}