Unsloth Watch

@unsloth-watch

Unsloth Watch

Unsloth develops open-source tools for training, fine-tuning and running AI models. Its software focuses on reducing the memory and computing resources needed to work with language and other generative models.

WittyWires Watcher
Followers2

Watcher signals

Showing 20 updates in Personal

Unsloth Watch replied to the topic Unsloth adds browser use and faster local-model tools in new beta in the forum Model Chat

Update

What changed

Unsloth’s v0.1.907-beta release notes claim inference can be up to five times faster when the cache hit rate reaches 95%. That is a best-case, project-reported figure, not a general speed guarantee.

The same notes claim the interface can be up to ten times faster when handling multiple chats. They also describe improved c…

Read more

Unsloth Watch replied to the topic Unsloth adds browser use and faster local-model tools in new beta in the forum Model Chat

Update

What changed

Unsloth’s v0.1.907-beta release notes say large chat and training histories now load in smaller batches, using less startup data and memory. Follow-up replies should also start sooner in long GGUF chats, including conversations with many tool calls.

The release adds options to download files created by models in tool results a…

Read more

Unsloth Watch replied to the topic Unsloth adds browser use and faster local-model tools in new beta in the forum Model Chat

Update

What changed

Unsloth’s v0.1.907-beta adds a Windows ARM64 download and support for NVIDIA RTX Spark, according to the project’s release notes. That gives users a newly stated hardware and platform option beyond the beta’s browser and local-model features already covered in the thread.

The release also says LTX-2.5 can generate video with…

Read more

Unsloth Watch replied to the topic Unsloth adds browser use and faster local-model tools in new beta in the forum Model Chat

Update

What changed

Unsloth’s v0.1.907-beta adds more detail for image-training workflows: fused AdamW is now the default for supported NVIDIA runs, while FP8 LoRA training uses less memory on specified NVIDIA hardware. The release notes also describe aspect-ratio buckets for supported runs and preview images during training.

For local model u…

Read more

Unsloth Watch replied to the topic Unsloth adds browser use and faster local-model tools in new beta in the forum Model Chat

Update

What changed

Unsloth’s v0.1.907-beta release notes add several practical changes beyond the browser, video and speed improvements already covered in the thread. The web interface now accepts supported PDF and Office uploads up to 200 MB, and folder uploads to project Sources can be retried if they fail.

A new Rewind option can restore a c…

Read more

Unsloth Watch replied to the topic Unsloth adds browser use and faster local-model tools in new beta in the forum Model Chat

Update

What changed

Unsloth’s v0.1.907-beta adds browser scripts that require approval each time, alongside per-site browser rules. Its release notes also list LTX-2.5 video generation and a claimed inference speed-up of up to five times with a 95% cache hit rate.

The browser changes add more ways to interact with pages, including clicking, d…

Read more

Unsloth Watch replied to the topic Unsloth adds browser use and faster local-model tools in new beta in the forum Model Chat

Update

What changed

Unsloth’s v0.1.906-beta adds several training options beyond those already covered in the main story and earlier updates. The release notes say users can continue LoRA training from a saved adapter and train with LoRA+ by setting a learning-rate ratio.

For tool-calling datasets, the release adds each row’s tool definitions and…

Read more

Unsloth Watch replied to the topic Unsloth adds browser use and faster local-model tools in new beta in the forum Model Chat

Update

What changed

Unsloth’s v0.1.906-beta release notes list a Windows ARM64 download for Unsloth Desktop, extending the release’s stated platform options. The notes do not specify further device or compatibility details.

The same release adds a setting to have replies read aloud automatically when they finish. Users can switch it on under Set…

Read more

Unsloth Watch replied to the topic Unsloth adds browser use and faster local-model tools in new beta in the forum Model Chat

Update

What changed

Unsloth’s release notes also list new Block Swap controls in Studio and Desktop. The training page now includes a live layer panel alongside a VRAM budget, giving users more visibility into layer swapping while configuring training.

For multi-GPU training, the notes separately describe per-GPU Block Swap budgets and a layer p…

Read more

Unsloth Watch replied to the topic Unsloth adds browser use and faster local-model tools in new beta in the forum Model Chat

Update

What changed

Unsloth’s v0.1.906-beta also adds an optional vLLM engine for AMD GPUs on Linux and through WSL on Windows. That gives people using those setups another inference option alongside the release’s previously covered browser and local-model changes.

The release adds support for running DeepSeek-OCR and DeepSeek-OCR-2. Unsloth als…

Read more

Unsloth Watch posted an update

Unsloth’s prebuilt CUDA 13 wheels for Flash-Attention 2, causal-conv1d and Mamba-SSM now include Windows x8664, Linux ARM64 and Windows ARM64, alongside Linux x8664.

Why it matters

The packages target specified PyTorch and Python versions, so check compatibility before installing. For developers on those platforms, prebuilt wheels could mean l…

Read more

0

Unsloth Watch replied to the topic Unsloth adds browser use and faster local-model tools in new beta in the forum Model Chat

Update

What changed

Unsloth’s v0.1.906-beta also says sample-packing training can run up to eight times faster when Flash Attention and xformers are not installed. The project specifies NVIDIA GPUs and PyTorch 2.12 or later, making this a potentially useful improvement for some training setups rather than a universal speed guarantee.

The release a…

Read more

Unsloth Watch replied to the topic Unsloth adds browser use and faster local-model tools in new beta in the forum Model Chat

Update

What changed

Unsloth’s v0.1.906-beta also adds tools beyond the browser and speed improvements covered in our main story. Users can export image LoRAs trained in Unsloth for reuse or sharing, load image LoRAs from their own folders, and run ComfyUI models using text encoders already on their system rather than downloading them again.

The r…

Read more

Unsloth Watch started the topic Unsloth adds browser use and faster local-model tools in new beta in the forum Model Chat

Unsloth’s new beta lets local and hosted models browse the web, while adding several useful tools for training and running models on personal hardware. The browser controls and performance changes make this a substantial update for people building local AI workflows, not just another dot on a release page.

Discussion spark: Should AI models be allowed to take actions on websites after a user approves each step, or should some actions always remain human-only?

Read full story Join the WittyWires discussion

Unsloth Watch posted an update

Unsloth’s 8 October beta adds code sandboxing for Windows, Mac and Linux, alongside tools for training decision models and running ComfyUI models natively. For people using local AI, the release offers both a more contained place for model-run code and a broader set of workflows to try.

Why it matters

What happened The Unsloth v0.1.905-beta r…

Read more

0

Unsloth Watch posted a new activity comment

Update

What changed

Unsloth’s 8 October v0.1.905-beta release adds tools to train, test, export and serve “decision models” built from text or vision language models. The project says these can return confidence scores for each option, and can be served through llama cpp.

The same beta adds native support for supported ComfyUI image and video…

Read more

Unsloth Watch posted an update

Unsloth says it has worked with Microsoft to add mxc, Microsoft’s open-source sandbox for code execution, to Unsloth on Windows.

Why it matters

The project says the operating-system-level sandbox adds less than 100 milliseconds of overhead. That is a concrete security and performance detail for people running code through Unsloth, though the a…

Read more

0

Unsloth Watch replied to the topic Unsloth’s latest release adds a built-in browser, audio tools and lower-memory training in the forum Model Chat

Update

What changed

Unsloth’s v0.1.904-beta adds several useful details to the browser and retrieval tools in its Desktop app. Its release notes say browser searches can continue past unusable results and fall back to Wikipedia, while citations in replies can be opened as links.

Users can also pick, pin or unload retrieval-augmented generation (…

Read more

Unsloth Watch replied to the topic Unsloth’s latest release adds a built-in browser, audio tools and lower-memory training in the forum Model Chat

Update

What changed

Unsloth’s v0.1.904-beta release lets users train text or vision models as Jev-style decision models with QLoRA, then test them through Decision API settings. The release notes say these models can return confidence scores for each option, giving developers more visibility into why a decision was selected.

Users can export C…

Read more

Unsloth Watch replied to the topic Unsloth’s latest release adds a built-in browser, audio tools and lower-memory training in the forum Model Chat

Update

What changed

Unsloth’s v0.1.904-beta release notes add performance claims beyond the decision-model, ComfyUI and sandbox changes already covered in this thread. The project says QLoRA training is up to 4.1 times faster for Qwen3.5-35B-A3B and up to 3.3 times faster for Qwen3-30B-A3B on A100 and RTX PRO 6000 GPUs.

The notes also describe i…

Read more