CVE-2026-124994: Turboderp-org ExLlamaV3 NGramEmbedding ngram.cu ngram_dequant out-of-bounds
A security flaw has been discovered in Turboderp-org ExLlamaV3 up to 1.6.3. The affected element is the function ngramdequant of the file exllamav3/exllamav3ext/ngram.cu of the component NGramEmbedding. The manipulation of the argument headbias results in out-of-bounds read. The attack may be launched remotely. The exploit has been released to the public and may be used for attacks. Upgrading to version 1.6.4 is sufficient to fix this issue. The patch is identified as c9a57ae832a3035e264e0a412e1daf4ebb699b98. It is suggested to upgrade the affected component.
Affected Software
Remediation
Recommended actions to resolve this vulnerability, in priority order.
- Upgrade
Upgrade
Turboderp-org ExLlamaV3to a version that resolves this vulnerability.Fixed in 1.6.4Patch c9a57ae832a3035e264e0a412e1daf4ebb699b98
Event History
Frequently Asked Questions
Which deployments should be considered affected?
Deployments using Turboderp-org ExLlamaV3 version 1.6.3 or earlier are affected. The vulnerable code is in the NGramEmbedding component's ngram_dequant function.
What conditions does exploitation require?
The issue can be attacked remotely by manipulating the head_bias argument. The CVSS vector indicates no privileges are required, but user interaction is required.
How should teams prioritize remediation?
A public exploit has been released, and the issue can affect confidentiality, integrity, and availability. Upgrade to version 1.6.4, which is identified as sufficient to fix the issue.
How can I verify that the fix is present?
Confirm that ExLlamaV3 has been upgraded to version 1.6.4 or later. The referenced patch is c9a57ae832a3035e264e0a412e1daf4ebb699b98.