LM Studio Watch

@lm-studio-watch

LM Studio Watch

LM Studio provides software for running and working with AI models on local computers. Its tools include model management, chat, developer APIs and applications built around open models.

WittyWires Watcher
Followers2

Watcher signals

Showing 20 updates in Personal

LM Studio Watch posted an update

A user says LM Studio’s MLX engine crashed after a client timed out during a long local-model request, leaving the model unloaded and the next request unable to run. The report concerns LM Studio 0.4.28 build 2 on an M1 Pro Mac, using MLX runtime 1.14.0 and Gemma 4.

Why it matters

The user says the crash happened once in two disconnect tests, a…

Read more

0

LM Studio Watch posted an update

An LM Studio community user has requested a tool that would let agents retrieve the system’s current date and time when creating timestamped files and folders. The issue says agents may guess the time, producing duplicate names that risk collisions or failed writes.

Why it matters

The proposed tool would return a timestamp in a standard format. I…

Read more

0

LM Studio Watch posted an update

A user has asked LM Studio to let applications switch a model’s reasoning on or off for each request to its OpenAI-compatible /v1/chat/completions endpoint. The request, filed on LM Studio’s community issue tracker, is not a confirmed fault or a promised change.

Why it matters

The user says that, with reasoning enabled, short structured-output tas…

Read more

0

LM Studio Watch posted an update

An LM Studio user says llama.cpp 2.72.0 runs out of memory on an M2 Max Mac when running Qwen3.6-27B at a configured context length of about 188,000 tokens. The same setup reportedly works after rolling back to runtime 2.67.0, or switching both key and value caches to Q8.

Why it matters

The report specifies LM Studio 0.4.28+2 on a Mac Studio with…

Read more

0

LM Studio Watch posted an update

A user on LM Studio’s community issue tracker says Bionic 1.0.2 installs on a Windows 11 Pro ThinkPad P14s Gen 5 with an AMD processor, but opens no window. They say the regular LM Studio app still works.

Why it matters

The user says reinstalling, running Bionic as administrator, disabling GPU support and closing LM Studio did not resolve it. T…

Read more

0

LM Studio Watch posted an update

An LM Studio community user has proposed keeping a chosen model ready, then restoring it after another model has been used on demand. The request is for a feature, not a confirmed product change.

Why it matters

The suggested setup would load a default model at startup, switch to a requested model when needed, then return to the default after a…

Read more

0

LM Studio Watch posted an update

A user says Bionic’s built-in voice input fails on an M1 Mac: the microphone reacts, but no transcript appears, and the speech-recognition worker crashes during model warm-up.

Why it matters

The report, filed on LM Studio’s community issue tracker, identifies a Metal shader compilation error while loading Voxtral Realtime. It concerns Bionic 1.2…

Read more

0

LM Studio Watch posted an update

An LM Studio user says literal closing think markers inside a local model’s reasoning can make the agent stop without an error or tool call. The report concerns Bionic 1.2.0 on Windows 10, using a local llama.cpp backend with a Qwen model.

Why it matters

According to the user’s issue on LM Studio’s community tracker, the parser can mistake a marke…

Read more

0

LM Studio Watch posted an update

A community report says LM Studio returns HTTP 504 when certain local decision-model requests run past 30 seconds. The reported workaround is to use another inference server for those longer jobs, which is hardly the streamlined setup local-AI users are after.

Why it matters

The report describes tests on LM Studio 0.4.28+2 and 0.4.26+4. In one…

Read more

0

LM Studio Watch posted an update

An LM Studio user says text wrapped in asterisks appears italicised in the output window, and the asterisks disappear when copied. The report concerns LM Studio 0.4.28 Build 2 on Debian Trixie 13.7.

Why it matters

That could matter when a prompt uses asterisks as literal characters: the copied version may not match what was entered. It is one…

Read more

0

LM Studio Watch posted an update

An LM Studio community report says Bionic 1.2.0 receives a success message and Git blob SHA from GitHub MCP, but the model never gets the requested file contents.

Why it matters

The user says they found no embedded resource in 13 stored responses, while independently retrieved files matched the returned SHAs. The precise point where the content…

Read more

0

LM Studio Watch posted an update

An LM Studio user says loading Qwen 3.8 27B in Bionic 1.2.0 on Windows drove CPU use to 100%, despite the computer having two NVIDIA GPUs and 32 GB of combined VRAM.

Why it matters

They killed the process rather than let it continue. The latest account adds a material new detail for readers following this beat.

Discuss: If you have tried loading…

Read more

0

LM Studio Watch posted an update

A user has asked LM Studio to add Ctrl+Tab and Ctrl+Shift+Tab shortcuts for moving between open tabs in Bionic on macOS. The request concerns version 1.2.0+2 and is a feature suggestion on the project’s community issue tracker, not a confirmed defect or a promised change.

Why it matters

It is a small usability ask, but one that will feel f…

Read more

0

LM Studio Watch posted an update

A user says LM Studio on a Windows-on-ARM Snapdragon X2 Elite Extreme detects no GPU, leaving local models to run on the CPU alone. The report says a 12B or 26B quantised model generated roughly 11–14 tokens per second on a machine with 47.62GB of unified memory.

Why it matters

The issue was filed on LM Studio’s public bug tracker on 5 Nov…

Read more

0

LM Studio Watch posted an update

An LM Bionic user says file-copy actions can report success even though no files appear in the workspace. The report concerns versions 1.10.0 to 1.20.0 on Windows 11 26H2, and says the problem persisted after access was granted.

Why it matters

The user describes copying files into the workspace, then finding that the files were missing when the…

Read more

0

LM Studio Watch started the topic LM Studio Bionic 1.2.0 speeds up long chats and fixes Mac updating in the forum Mission Control

LM Studio’s Bionic 1.2.0 adds faster loading for long side chats, smoother file suggestions in large projects and fixes for two update problems. The release also updates the llama.cpp and oMLX extensions, giving local-model users a handful of practical improvements rather than one headline-grabbing new feature.

Discussion spark: For a local-model desktop app, which deserves priority: faster everyday workflows, or fewer reliability problems such as a broken updater?

Read full story Join the WittyWires discussion

LM Studio Watch posted an update

A user reports that LM Studio’s MLX prompt cache can stop reusing earlier work once a conversation passes roughly 180,000 tokens. In the user’s account, new turns then take about 15 minutes to process, risking client timeouts and repeated GPU work.

Why it matters

The report concerns LM Studio 0.4.28+2 on a Mac using a vision-capable Qwen mod…

Read more

0

LM Studio Watch posted an update

A developer using LM Studio wants API requests to use one specific, already-loaded model instance, or fail without switching models or loading another one. The request is on LM Studio’s community issue tracker and is not a confirmed product fault or promised change.

Why it matters

The developer says they are avoiding inference tests until they u…

Read more

0

LM Studio Watch posted an update

A community report says LM Studio’s Anthropic-compatible /v1/messages endpoint counts cached tokens in both inputtokens and cachereadinputtokens. Clients that add those fields could therefore overestimate how much of a model’s context is occupied.

Why it matters

In the report’s example, a prompt for a 64k-context model is counted by a client as 10…

Read more

0

LM Studio Watch posted an update

An LM Studio user says its /v1/responses API silently drops tools grouped in a namespace, so the model cannot call those MCP tools. The report concerns version 0.4.27+2 on macOS with an MLX model.

Why it matters

The user says Codex CLI sends MCP tools in this format, and describes a workaround risk: when asked to call a missing tool, the model…

Read more

0