Release 0.5.4

0.5.4 (2025-02-18)

Bug Fixes

• android: skip library load on not supported arch (bb9d6e8)
• ts: add missing tool_calls in NativeCompletionResult (6572fc7)

Features

• ios: add no_gpu_devices param (dbca475)
• mock: update return of initContext (64b02c4)
• sync llama.cpp (#122) (2b1a8bd)

Release 0.5.3

0.5.3 (2025-02-07)

Bug Fixes

• android: add missing tool_calls parse (f671500)
• ios: catch tool_calls parse (b2a0e2d)
• ts: json_object type of response_format (01cd6a2)
• ts: return type in native getFormattedChat & update mock (8e76b38)

Release 0.5.2

0.5.2 (2025-02-06)

Bug Fixes

• ios: error on set toolCalls on interrupt (91913f1)

Features

• add chat_template for override default on init (6a6a9bd)
• implement methods to listen native logs (#119) (38858a7)
• ios: still enable metal backend on simulator & build metallib for simulator (#120) (94ac13e)
• sync llama.cpp (f3887dc)

Release 0.5.1

0.5.1 (2025-02-04)

Bug Fixes

• ios: catch std::exception for getFormattedChat (c36b085)

Features

• add response_format completion param (#118) (c39a599)
• ios, tvos: use residency sets (82bb747)
• ios: enable GGML_METAL_USE_BF16 (9754170)
• sync llama cpp & support use jinja template / tool_calls (#117) (dd2179e)

Release 0.5.0

0.5.0 (2025-01-24)

Features

• add nEmbd in model details (781b49a)
• android: add loaded lib name in context (9f63709)
• default context limit to unlimited (3f88f6e)
• make prebuilt libs (#112) (85bceef)
• sync llama.cpp (#110) (7e56a2b)
• ts: update mock (263473a)

Release 0.4.8

0.4.8 (2025-01-09)

Bug Fixes

• log: implement Android logging in rn-llama.hpp as opposed to printf (#106) (fb3896e)

Features

• android: enable runtime repacking for Q4_0 quantization on aarch64 (#105) (758157b)
• expose n_ubatch and dynamically adjust ntokens for bench (#104) (9c25ec4)
• mock: add model metadata & mock data for tokenize / embedding (3296388)
• sync llama.cpp (#108) (b539012)

Release 0.4.5

0.4.5 (2024-12-23)

Features

• sync llama.cpp (#100) (951436c)

Release 0.4.4

0.4.4 (2024-12-20)

Bug Fixes

• cpp: progress_callback patch (#99) (b8d0f20)
• ios: newarch setup (0eab36a), closes #96
• ts: missing mock methods (a1d5413)

Release 0.4.2

0.4.2 (2024-11-21)

Bug Fixes

• ts, docs: missing docs of classes / types (1bfc8d8)

Features

• expose DRY sampler params (#91) (0c04b5e)
• support multiple lora files & dynamic apply / remove lora (#92) (9b6ea9f)
• sync llama.cpp (#95) (276a90a)

Release 0.4.1

0.4.1 (2024-11-17)

Bug Fixes

• cpp, ios: add prefix for iq2/iq3 func for avoid redef from another libs used ggml (8ce5216)
• cpp: validateModelChatTemplate missing messages check (c1d15a3)
• ts: remove file:// prefix for lora param (da3e9a7)

Features

• add static method for read model info from gguf (#87) (8bf9dd8)
• cpp: remove unused json and json-schema-to-grammar in common (#90) (0a37cda)
• expose flash_attn / cache_type_k / cache_type_v (4ce8ff8)
• no longer need disable mmap for lora (38fa660)
• sync llama.cpp (#89) (5e1b30a)
• ts: expose template for getFormattedChat (c473995), closes #84
• update embedding method (#88) (6190f57)