v0.9.5

Changes:

• llama.cpp updated to b2135
• Added the ability to download models from the application menu
• Added progress indicator for model loading
• Fixed disappearing keyboard bug
• Fixed some other bugs

v0.9.2

Changes:

• Added possibility to specify System Prompt, which will be added to the text of the first message in the session. See FAQ.
• Added ability to clone chat (without message history)
• Chats are sorted by last modification date
• Clear chat history button is placed on the toolbox.
• You can now use both {prompt} and {{prompt}} designations in templates
• Fixed a bug with displaying already deleted chats and models
• Fixed some other bugs
• Templates have been updated

v0.9.0

Changes:

• llama.cpp updated to b1891
• added support for Phi2, TinyLlama and other models
• various GUI improvements
• added clear chat button
• user templates improve
• fixed token to string issue
• fixed gpt2 metal
• fixed many other errors

v0.8.1

Changes:

• fixed autoscroll

** Metal support temporary disabled for GPT2 models
** More about LoRA here https://github.com/guinmoon/LLMFarm/blob/main/lora.md

v0.8.0

Changes:

• llama.cpp updated to b1601
• added support for StableLM-3b-4e1t models
• added support for Qwen models
• added the possibility to merge LoRA with the model
• added merge and train LoRA progress bar
• added the possibility to save user templates
• added multyline input
• fixed many other errors

** Metal support temporary disabled for GPT2 models
** More about LoRA here https://github.com/guinmoon/LLMFarm/blob/main/lora.md

v0.7.5

Changes:

• added LoRA train support experimental
• add BOS/EOS token to begin/end of prompt options
• handle special tokens options
• model loading indication
• token/sec indicator on message
• fixed some errors

** Due to high RAM consumption LoRA training on iPhone is possible only on Pro models
It is recommended to use q8_0 quantization for LoRA training.

v0.7.0.1

Changes:

• added special token support for prompt template like <s>User: {{prompt}}</s>
• fixed tokenizer bug that could cause application crashing

v0.7.0

Changes:

• llama.cpp updated to b1396
• added support for MPT models
• added support for Bloom models
• added support Metal for q5_0, q5_1 quantization
• gpt-2 now with Metal support
• LoRA adapters support (More about LoRA here)
• fixed mirostat for non llama
• fixed premature completion of predictions
• fixed many other errors

v0.6.2

Changes

• starcoder (santacoder) with Metal and mmap support (GGUF)
• infinite text generation by reseting n_past
• fixed some errors
• fixed mmap always false

v0.6.1

Changes

• fix crash on n_tokens > context size
• fix llama mlock