0.10.0

Features

• redirected stability.stable-diffusion-xl to stability.stable-diffusion-xl-v1, since stability.stable-diffusion-xl-v0 is deprecated (#92)
• supported Claude 3 Opus (#90)

Fixes

• fixed bug with ignored max_tokens in Claude 3 (#99)

Other

• bump aiohttp from 3.9.2 to 3.9.4 (#88)
• bump pydantic from 1.10.12 to 1.10.13 (#95)

0.9.0

Features

• migrated to the latest SDK and thus supported new way of OTEL configuration (#85)
• removed deprecated claude-v1 model (#68)
• supported Claude 3 Haiku model with deployment name anthropic.claude-3-haiku-20240307-v1:0 (#84)

Fixes

• deprecated DIAL_USE_FILE_STORAGE env var (breaking changes, see the issue for a migration guide: #73)

Other

• bump idna from 3.4 to 3.7 (#87)

0.8.0

Features

• supported Claude 3 Sonnet (#77)
• supported tokenize and truncate_prompt endpoints (#82)

Fixes

• DEFAULT_REGION env var is marked as deprecated in favour of AWS_DEFAULT_REGION env var (#80)

Other

• bump black from 23.3.0 to 24.3.0 (#79)
• bump epam/ai-dial-ci from 1.7.0 to 1.8.0 (#78)
• bump langchain-core from 0.1.29 to 0.1.30 (#76)
• bump langchain-core from 0.1.30 to 0.1.35 (#83)
• bump peter-evans/slash-command-dispatch from 3.0.1 to 4.0.0 (#75)
• update github workflows (#74)

0.7.0

Fixes

• migrated to sdk==0.7.0: discarded_messages response field turned into a list of indices (#72)

Other

• migrated to openai>=1.0 and langchain-openai libs (#71)

0.6.0

Features

• Claude V2.1: supported array and enum types in the tools schema (#61)

Fixes

• file metadata definition adjusted to follow DIAL Core 0.7.0 API (#60)
• Claude V2.1: the completion is streamed in a single chunk when tools are used (#62)

Other

• GitHub workflow update (#64)
• bump aidial-sdk from 0.6.0 to 0.6.2 (#66)
• bump aiohttp from 3.9.0 to 3.9.2 (#63)
• bump jinja2 from 3.1.2 to 3.1.3 (#53)

0.5.0

Features:

• added OTLP export (#52)
• configure service name via env var (#55)

Fixes:

• avoid generating empty content for Stability (#56)
• capitalized name of stage with image (#51)
• fixed stability adapter (#48)
• propagating per request key (#57)

Other:

• chore: Setup excludes for OSS Review Toolkit (#50)

0.4.1

Fixes:

• fixed stability adapter (#48) (#49)

0.4.0

Features:

• supported tools/functions in claude 2.1 (#43)

0.3.0

Features:

• added llama 2 and cohere models (#30) (#35)
• added pseudo chat config; added wrapper for Bedrock client (#36)
• migrated to new auth method (#44)
• saving stability artifacts to DIAL file storage (#34)
• supported turn-based dialogue partitioning for llama model (#39)

Fixes:

• added missing dot before file extension (#47)
• don't use file storage by default; renamed env vars (#37)
• fixed double extension bug (#46)

Other:

• chore: add e2e tests trigger (#28)
• chore: bump aiohttp from 3.8.6 to 3.9.0 (#32)

0.2.0

Features:

• added integration tests for max_tokens and stop sequence parameters (#23)
• allowed empty messages in each language model (#23)
• supported history truncation via max_prompt_tokens and discarded_messages parameters (#23)
• supported streaming for Titan and Claude (#23)

Fixes:

• used number of bytes as token count estimator for AI21 and AWS Titan (#23)
• fixed AI21 temperature setting (#23)
• removed Assistant prefix occasionally generated by Titan (#23)