最後更新: 2024-09-26
前言
只記錄了在 Linux 版上的重大亮點
目錄
Ollama Changelog Summary
v0.3.12
- New models: Qwen 2.5 Coder(7b, 1.5b), Llama 3.2(1b, 3b)
v0.3.11
- New models: Qwen 2.5, Reader-LM(HTML->Markdown), Mistral-Small(22B)
v0.3.10
- New models: DeepSeek-V2.5(236b{133GiB}), MiniCPM-V(vision), Yi-Coder(9b, 1.5b)
v0.3.8 ~ v0.3.9
- bugfix
v0.3.7
- 加入了 Phi 3.5 (3.8b)
- binary 及 libraries 打包成 .tgz, size 成 1.3G
v0.3.6 ~ v0.3.2
v0.3.1
- Added support for min_p
v0.3.0
-
supports tool calling
- Functions and APIs
- Web browsing
- Code interpreter
v0.2.8 ~ v0.2.1
v0.2.0
- Parallel requests (using only a little bit of additional memory for each request)
情景: Handling multiple chat sessions at the same time
- Multiple models (loading different models at the same time)
ollama ps
- New models: GLM-4, CodeGeeX4
v0.1.48
- bugfix
v0.1.47
- New models - Gemma 2 (9B and 27B)
v0.1.46
- Increased model loading speed with ollama run, especially if running an already-loaded model
- Improved model loading times when models would not completely fit in system memory on Linux
v0.1.45
- New models: DeepSeek-Coder-V2
v0.1.42
- New models: Qwen 2
v0.1.40
- New models: Codestral, IBM Granite Code
v0.1.39
- New models - Phi-3 Mini 128K and Phi-3 Medium 128K
- Added a Ctrl+W shortcut to ollama run
v0.1.38
- ollama ps # displays currently loaded models, their memory footprint, and
# the processors used (GPU or CPU):
- /clear # To clear the chat history for a session when running ollama run
v0.1.35
- 當 importing 時可以修改 models 的 quantize (--quantize, -q)
ollama create -f Modelfile --quantize q4_0 mymodel
- Ctrl+J characters will now properly add newlines in ollama run
OpenWebUI Changelog Summary
v0.3.32
- Bug Fix
v0.3.31
- code blocks now allow live editing directly in the LLM response, with live reloads supported by artifacts.
-
New floating buttons appear when text is highlighted in LLM responses,
offering deeper interactions like "Ask a Question" or "Explain". - Implemented lazy loading of large dependencies to minimize initial memory usage, boosting performance.
- Expandable Content Markdown Support
v0.3.25~v0.3.30
- Bug Fix
v0.3.24
- Users can now mark responses as favorite directly from the chat overview
- Create Message Pairs with Shortcut: Ctrl+Shift+Enter
- Expanded User Prompt Variables: weekday, timezone, and language
- support for 'audio/x-m4a' files
- PDF citations now open at the associated page, streamlining reference checks
v0.3.23
- 無加亮點
v0.3.22
- Multiple Vector DB Support: Milvus
v0.3.21
- Enabled /api/embed endpoint proxy support.
- Now displays the total number of documents directly within the dashboard.
v0.3.19 ~ v0.3.20
- Bug Fix
v0.3.18
- Direct Database Execution(Python) for Tools & Functions
v0.3.16 - 2024-08-27
- Migrated configuration handling from config.json to the database.
- allowing users to click to copy content from code spans directly.
v0.3.15 - 2024-08-21
-
Temporary Chat Activation
chat sessions directly through the URL (URL parameter 'temporary-chat=true')
v0.3.14 - 2024-08-21
- 當同時問兩個不同 models 後, 可以 Merging 它們的回應為一個
-
Enhanced Shift key quick actions for hiding/unhiding and deleting models
(Workspace > Models 按 shift key 可以改變 action) - User messages are now rendered in Markdown
v0.3.13 - 2024-08-14
- Auto-Install Tools & Functions Python Dependencies
-
Websocket Reconnection
(reconnect when a websocket is closed)
v0.3.12
- 無加亮點
v0.3.11
- Added 'Min P' parameter in the advanced settings for customized model precision control.
v0.3.10
- 無加亮點
v0.3.9
- A new "Action" function to write custom buttons to the message toolbar.
v0.3.8
- Darker OLED Theme
v0.3.7
- 無加亮點
v0.3.6
- "Functions" Feature
- Files API