Releases: epam/ai-dial-adapter-bedrock
Releases · epam/ai-dial-adapter-bedrock
0.16.0
0.15.1
0.15.0
0.14.0
Features
- supported multiple AWS accounts and regions within a single bedrock adapter deployment (#141)
- added Llama 3.1 8B/70B/405B models (#136)
- added Cohere embedding models (Embed English and Embed Multilingual) (#132)
- added
code=truncate_prompt_error
to errors raised by prompt truncation algorithm (#134)
Fixes
- supported zero-argument tools/functions (#134)
- Claude 3 fails when
request.max_prompt_tokens
is provided instead of ignoring it (#144) - improved reporting of content filter errors for Stability (#143)
- supported passing through of 429 errors reported by Bedrock (#138)
- image tokens are accounted in token usage computation for Titan Multimodal embeddings (f5d734f)
Other
0.13.3
0.13.2
0.13.1
0.13.0
Features
- supported native tool use in Claude V3 chat model (#116)
- added Claude 3.5 Sonnet chat model (#117)
- added Titan Multimodal Embeddings Generation 1 (G1) (#114)
- added Amazon Titan Text Embeddings V2 (#112)
- added Titan Embeddings G1 – Text v1.2 (#113)
Other
- bump urllib3 from 1.26.18 to 1.26.19 (#115)
0.12.0
0.11.0
Features
- supported Llama 3 8B and Llama 3 70B models (#102)
- supported legacy tool use for Claude 3 models (#98)
- added tests for function/tool usage; migrated tests from langchain to openai; removed langchain dependency (#100)
- reporting user-facing error messages via
display_message
(#89)
Fixes
- tool call ids made unique (#107)