Gemma 3 28b crashes as soon as an image is provided in all versions after 0.3.21 (Build 4)

[chris114782]

Which version of LM Studio?
LM Studio 0.3.25 (Build 2)

Which operating system?
Arch Linux (Using an Nvidia 4080 Super w/16GB VRAM)

What is the bug?
In every version after 0.3.21 (Build 4) Gemma 3 28b has been crashing as soon as an image is provided. Running 0.3.21 (Build 4) still works flawlessly.

Other vision models are seemingly OK, this only seems to impact Gemma 3 28b, but I'm not sure if this is a model issue or an LM studio issue, as the older build runs fine.

Screenshots

Logs

[ModelLoadingProvider] Requested to load model google/gemma-3-27b with opts {
  identifier: { desired: 'google/gemma-3-27b', conflictBehavior: 'bump' },
  instanceLoadTimeConfig: { fields: [] },
  instanceOperationTimeConfig: { fields: [] }
}
[CachedFileDataProvider] Watching file at /home/chris/.lmstudio/.internal/user-concrete-model-default-config/google/gemma-3-27b.json
[ModelLoadingProvider] Started loading model google/gemma-3-27b
[ModelProxyObject(id=google/gemma-3-27b)] Forking LLMWorker with custom envVars: {"LD_LIBRARY_PATH":"/home/chris/.lmstudio/extensions/backends/vendor/linux-llama-cuda12-vendor-v1","CUDA_VISIBLE_DEVICES":"0"}
[ProcessForkingProvider][NodeProcessForker] Spawned process 6856
[LMSInternal][Client=LM Studio][Endpoint=getOrLoad] Requested get or load model: google/gemma-3-27b
D [LMSExternal] [Client=plugin:installed:lmstudio/rag-v1] Client created.
[InternalPluginsProvider][PluginProcess(lmstudio/rag-v1)] stdout: [PromptPreprocessor] Register with LM Studio

D [LMSExternal] [Client=plugin:installed:lmstudio/rag-v1][Endpoint=setPromptPreprocessor] Registering promptPreprocessor.
[InternalPluginsProvider][PluginInstance(id=3,identifier=lmstudio/rag-v1)] Added plugin handle 0. # of references: 1.
[ConversationsProvider] Creating new working directory: /home/chris/.lmstudio/working-directories/1757019759393 for 1757019759393.conversation.json
[LMSInternal][Client=LM Studio][Endpoint=applyPromptTemplate] Applying prompt template for specifier: {"type":"instanceReference","instanceReference":"sjJOhnmudi/XafsBKhMqUr3t"}
[ConversationsProvider] Removing empty working directory: /home/chris/.lmstudio/working-directories/1757019759393 for 1757019759393.conversation.json
22:04:22.350 › (node:6438) [DEP0147] DeprecationWarning: In future versions of Node.js, fs.rmdir(path, { recursive: true }) will be removed. Use fs.rm(path, { recursive: true }) instead
(Use `lm-studio --trace-deprecation ...` to show where the warning was created)
[InternalPluginsProvider][PluginInstance(id=3,identifier=lmstudio/rag-v1)] Disposed plugin handle 0. # of references: 0.
[ConversationsProvider] Creating new working directory: /home/chris/.lmstudio/working-directories/1757019759393 for 1757019759393.conversation.json
[InternalPluginsProvider][PluginInstance(id=3,identifier=lmstudio/rag-v1)] Added plugin handle 1. # of references: 1.
[InternalPluginsProvider][PluginProcess(lmstudio/rag-v1)] stdout: [PromptPreprocessor][Request (uxaYjz)] New preprocess request received.

[InternalPluginsProvider][PluginProcess(lmstudio/rag-v1)] stdout: [PromptPreprocessor][Request (uxaYjz)] Preprocess request completed.

[InternalPluginsProvider][PluginInstance(id=0,identifier=lmstudio/default-prediction-loop-handler)] Added plugin handle 2. # of references: 1.
[   PredictionLoopHandler][Request (9MXSw5)] New prediction loop handling request received.
[LMSInternal][Client=plugin:builtin:lmstudio/default-prediction-loop-handler][Endpoint=getOrLoad] Requested get or load model: google/gemma-3-27b
[ProcessForkingProvider][NodeProcessForker] Exited process 6856
22:04:24.604 › [LMSInternal][Client=plugin:builtin:lmstudio/default-prediction-loop-handler][Endpoint=predict] Error in channel handler: Error: The model has crashed without additional information. (Exit code: null)
    at _0xed8f48._0x1e82cc (/tmp/.mount_lm-stupyfceb/resources/app/.webpack/main/index.js:101:13505)
    at _0xed8f48.emit (node:events:518:28)
    at _0xed8f48.onChildExit (/tmp/.mount_lm-stupyfceb/resources/app/.webpack/main/index.js:91:206261)
    at _0x931c63.<anonymous> (/tmp/.mount_lm-stupyfceb/resources/app/.webpack/main/index.js:91:205616)
    at _0x931c63.emit (node:events:530:35)
    at ChildProcess.<anonymous> (/tmp/.mount_lm-stupyfceb/resources/app/.webpack/main/index.js:477:20963)
    at ChildProcess.emit (node:events:518:28)
    at ChildProcess._handle.onexit (node:internal/child_process:293:12)
[LMSInternal][Client=plugin:builtin:lmstudio/default-prediction-loop-handler][Endpoint=predict] Canceled predicting due to channel error.
[   PredictionLoopHandler] Generation failed. Error: Channel Error
    at handle (/tmp/.mount_lm-stupyfceb/resources/app/.webpack/main/index.js:353:90020)
    at process.processTicksAndRejections (node:internal/process/task_queues:105:5)
- Caused By: Error: The model has crashed without additional information. (Exit code: null)
    at _0xed8f48._0x1e82cc (/tmp/.mount_lm-stupyfceb/resources/app/.webpack/main/index.js:101:13505)
    at _0xed8f48.emit (node:events:518:28)
    at _0xed8f48.onChildExit (/tmp/.mount_lm-stupyfceb/resources/app/.webpack/main/index.js:91:206261)
    at _0x931c63.<anonymous> (/tmp/.mount_lm-stupyfceb/resources/app/.webpack/main/index.js:91:205616)
    at _0x931c63.emit (node:events:530:35)
    at ChildProcess.<anonymous> (/tmp/.mount_lm-stupyfceb/resources/app/.webpack/main/index.js:477:20963)
    at ChildProcess.emit (node:events:518:28)
    at ChildProcess._handle.onexit (node:internal/child_process:293:12) {
  title: 'The model has crashed without additional information. (Exit code: null)'
}
[InternalPluginsProvider][PluginInstance(id=3,identifier=lmstudio/rag-v1)] Disposed plugin handle 1. # of references: 0.
[ConversationsProvider] Removing empty working directory: /home/chris/.lmstudio/working-directories/1757019759393 for 1757019759393.conversation.json
22:04:24.607 › [LMSInternal][Client=LM Studio][Endpoint=predictNextMessage] Error in RPC handler: Error: Rehydrated error
crashed without additional information. (Exit code: null)
    at _0x58b844.<computed> (/tmp/.mount_lm-stupyfceb/resources/app/.webpack/main/index.js:720:132146)
    at _0x17548d.subscriber (/tmp/.mount_lm-stupyfceb/resources/app/.webpack/main/index.js:265:1528)
    at _0x17548d.notifier (/tmp/.mount_lm-stupyfceb/resources/app/.webpack/main/index.js:550:183149)
    at process.processTicksAndRejections (node:internal/process/task_queues:105:5)
- Caused By: Error: Channel Error
    at handle (/tmp/.mount_lm-stupyfceb/resources/app/.webpack/main/index.js:353:90020)
    at process.processTicksAndRejections (node:internal/process/task_queues:105:5)
- Caused By: Error: The model has crashed without additional information. (Exit code: null)
    at _0xed8f48._0x1e82cc (/tmp/.mount_lm-stupyfceb/resources/app/.webpack/main/index.js:101:13505)
    at _0xed8f48.emit (node:events:518:28)
    at _0xed8f48.onChildExit (/tmp/.mount_lm-stupyfceb/resources/app/.webpack/main/index.js:91:206261)
    at _0x931c63.<anonymous> (/tmp/.mount_lm-stupyfceb/resources/app/.webpack/main/index.js:91:205616)
    at _0x931c63.emit (node:events:530:35)
    at ChildProcess.<anonymous> (/tmp/.mount_lm-stupyfceb/resources/app/.webpack/main/index.js:477:20963)
    at ChildProcess.emit (node:events:518:28)
    at ChildProcess._handle.onexit (node:internal/child_process:293:12)
[InternalPluginsProvider] Evicting plugin lmstudio/rag-v1.
[InternalPluginsProvider][PluginInstance(id=3,identifier=lmstudio/rag-v1)] Unload requested
D [LMSExternal] [Client=plugin:installed:lmstudio/rag-v1] Client disconnected.
[PluginsExternalAPIProvider][WsServer:AuthenticatedWsServer] Terminating connection because authentication was revoked
[InternalPluginsProvider][PluginInstance(id=3,identifier=lmstudio/rag-v1)] Unloaded

To Reproduce

1. Start a new chat
2. Load a vision model
3. Supply an image. My example image above is
   
4. Send the image to the model
5. crash

[neilmehta24]

Hey @chris114782 thank you for the report. Could you send us the following information please?

• Model debug info (right-click on the gear)

• Verbose logs during the crash (you'll have to enable verbose logging)

• A screenshot of your runtimes page in LM Studio (Ctrl + Shift + R)

[Sh1d0w]

Same issue for me

[vladikoles]

I have the same problem with google/gemma-3-12b on Ubuntu 22.04 (Intel Core i3-7100, 16GB RAM, nVidia P106-100 6GB).
Otherwise, this model works for me.
LM Srudio 0.3.25
LM Srudio 0.3.30 (Build 1)

[vladikoles]

I looked at the log.
It looks like there's not enough video memory for this operation.
I reduced the "GPU Offload" parameter value from 28 to 14.
And everything worked!