## unslothai/unsloth — v0.1.525-beta…v0.1.526-beta

_50 commits._

### Features
- **Studio: add a loaded models indicator with per-model eject (#8082)** (b3705f4)
- **Add Markdown conversation export (#7664)** (a3129df)
- **Add Model memory settings to keep a loaded model in VRAM (#8002)** (1b3c31e)

### Fixes
- **Fix on-device GGUF image/video picks asking for a quantization (#8149)** (7ff60be)
- **[Fix] Unsloth Studio: Strip ANSI escape codes from Studio tool output panes (#7976)** (051e71c)
- **Fix desktop image drops for chat attachments (#7982)** (f8497dc)
- **Studio: fix STT download progress and cache safety (#8035)** (3654a75)
- **fix(studio): harden training setup, lifecycle, and audio loading (#8103)** (0b147fd)
- **fix(AMD): let diffusion load on Windows ROCm, and say what a gated repo needs (#7981)** (6e18ebf)
- **studio: fix deep research progress reporting, loopback auth and model retries (#8129)** (d9e033a)
- **fix(security): re-baseline the huggingface-hub HTTP backoff loop for 1.x (#8135)** (4e19313)
- **perf(studio): make local model inventory scans linear (#8043)** (e96d020)

### Backend
- **Versioning** (36d836a)
- **Enforce the interface's video resolutions in the API too (#8159)** (56d410f)
- **Count a RECORD path claim before the row is filtered out (#8181)** (5f5f1e3)
- **Studio: hold the Mac capability verdict while an MLX repair can still overturn it (#8152)** (7103733)
- **Studio: route every Windows installer line through the UTF-8 stdout sink (#8148)** (07df950)
- **Pin the auto-load success toast by its branch, not by one spelling (#8183)** (502ec25)
- **Catch the llama-prebuilt validator stubs up with rocm_gfx (#8182)** (b0ea23b)
- **Do not stage diffusion files that are already in the cache (#8154)** (43c9efc)
- **Studio: exempt shared top-level trees from the integrity check for our own wheels too (#8179)** (c2c16b7)
- **Studio: bring back the sidebar scrollbar without losing row width (#8147)** (a8dec41)
- **Attach save_lora without a vLLM engine (#8127)** (21f8efd)
- **Studio: rate the API monitor Throughput tile on the decode window (#8139)** (90f2fca)
- **Render a static prefix when default_system_message is None (#8117)** (3177523)
- **Do not choose flex attention on a card that cannot run its kernel (#8137)** (e1db695)
- **Keep download scopes out of the GGUF quantization list (#8153)** (5c4623f)
- **Stop the CUDA test spoofs reporting an exhausted card (#8123)** (7e5df70)
- **Studio: surface API errors and align audio limits (#7946)** (bb2bab7)
- **Desktop: refuse to republish an existing release version (#7941)** (38dec58)
- **Publish studio_install_id without clobbering a concurrent writer (#8145)** (63bc5aa)
- **Bump the cargo-tauri group across 1 directory with 18 updates (#8107)** (61f6bee)
- **Deduplicate the scan-packages baseline, and gate it with tests (#8110)** (da72577)
- **Studio: retry Xet once on a stall before falling back to HTTP (#8166)** (d882da2)
- **Studio: stop research threads 409-ing on autosave, keep server-managed messages intact (#7956)** (9c2fda3)
- **Studio: fall back to CPU when auto-selected Vulkan crashes (#8050)** (738413a)
- **Pin the accelerator gates in the DiT precision preflight test (#8155)** (eab9c89)
- **Studio: per-request perf stats and slot usage in the API monitor (#8045)** (90d7a3c)
- **Report a denied UNSLOTH_LOCAL_LLAMA_CPP_DIR instead of aborting on errexit (#8143)** (9f37302)
- **Studio: keep row actions clear of overlay scrollbars (#8031)** (c84e3e2)
- **Desktop: ensure backend ownership for Tauri installs (#8034)** (fbf923e)
- **Reserve host RAM for AMD unified-memory APUs on the ROCm probe (#8131)** (6e0beb2)
- **Studio: check llama.cpp cache access before setup (#8032)** (69d555b)
- **Repair seven tests left behind by their own subjects (#8132)** (58cc1b2)
- **Studio: keep inference status probes off the event loop (#7944)** (47df5b8)
- **Desktop: block HTTP loopback requests from untrusted images (#8046)** (0ca37e5)
- **Clamp Deep Research max_tokens to loaded context window (#7985)** (1229fc5)
- **Bump install.sh / install.ps1 pin to unsloth>=2026.8.8 (#8114)** (483d751)
- **studio: expose llama-server batch and micro-batch sizes as first-class load settings (#7973)** (968134a)

### Chore
- **CI: prune superseded Actions caches (#8167)** (38cda28)

_Recap by [Repo Wrapped](https://repowrapped.com/gh/unslothai/unsloth?utm_source=github-action)._