v3.0 is here!

[github-actions[bot]]

✨ node-llama-cpp 3.0 is here! ✨

Read about the release in the blog post

---

3.0.0 (2024-09-24)

Features

• function calling (#139) (5fcdf9b)
• get embedding for text (#144) (4cf1fba)
• async model and context loading (#178) (315a3eb)
• token biases (#196) (3ad4494)
• automatic batching (#104) (4757af8)
• prompt completion engine (#225) (95f4645)
• model compatibility warnings (#225) (95f4645)
• Vulkan support (#171) (d161bcd)
• Windows on Arm prebuilt binary (#181) (f3b7f81)
• change the default log level to warn (#191) (b542b53)
• pull command (#214) (453c162)
• inspect gpu command (#175) (5a70576)
• inspect gguf command (#182) (35e6f50)
• inspect estimate command (#309) (4b3ad61)
• inspect measure command (#182) (35e6f50)
• init command to scaffold a new project from a template (with node-typescript and electron-typescript-react templates) (#217) (d6a0f43)
• move download, build and clear commands to be subcommands of a source command (#309) (4b3ad61)
• move seed option to the prompt level (#309) (4b3ad61)
• TemplateChatWrapper: custom history template for each message role (#309) (4b3ad61)
• Llama 3.1 support (#273) (e3e0994)
• Mistral chat wrapper (#309) (4b3ad61)
• Functionary v3 support (#309) (4b3ad61)
• Phi-3 support (#273) (e3e0994)
• extract all prebuilt binaries to external modules (#309) (4b3ad61)
• parallel function calling (#225) (95f4645)
• preload prompt (#225) (95f4645)
• onTextChunk option (#273) (e3e0994)
• flash attention (#264) (c2e322c)
• debug mode (#217) (d6a0f43)
• load LoRA adapters (#217) (d6a0f43)
• split gguf files support (#214) (453c162)
• stopOnAbortSignal and customStopTriggers on LlamaChat and LlamaChatSession (#214) (453c162)
• Llama 3 support (#205) (ef501f9)
• --gpu flag in generation CLI commands (#205) (ef501f9)
• specialTokens parameter on model.detokenize (#205) (ef501f9)
• interactively select a model from CLI commands (#191) (b542b53)
• automatically adapt to current free VRAM state (#182) (35e6f50)
• GGUF file metadata info on LlamaModel (#182) (35e6f50)
• use the tokenizer.chat_template header from the gguf file when available - use it to find a better specialized chat wrapper or use JinjaTemplateChatWrapper with it as a fallback (#182) (35e6f50)
• simplify generation CLI commands: chat, complete, infill (#182) (35e6f50)
• gguf parser (#168) (bcaab4f)
• use the best compute layer available by default (#175) (5a70576)
• more guardrails to prevent loading an incompatible prebuilt binary (#175) (5a70576)
• completion and infill (#164) (ede69c1)
• support configuring more options for getLlama when using "lastBuild" (#164) (ede69c1)
• get VRAM state (#161) (46235a2)
• chatWrapper getter on a LlamaChatSession (#161) (46235a2)
• minP support (#162) (47b476f)
• chat syntax aware context shifting (#139) (5fcdf9b)
• stateless LlamaChat (#139) (5fcdf9b)
• LlamaText util (#139) (5fcdf9b)
• show llama.cpp release in GitHub releases (#142) (36c779d)
• model metadata overrides (#273) (e3e0994)

BREAKING CHANGES

• completely new API

---

Shipped with llama.cpp release b3808

To use the latest llama.cpp release available, run npx -n node-llama-cpp source download --release latest. (learn more)

---

This discussion was created from the release v3.0.0.