0.17.2

Fixes

• Supported approximate tokenization of tools and functions (#170)

0.17.1

Fixes

• gpt-4o: fixed failing tool/function calling (#175, #176)
• removed Content-Encoding header from response to avoid decoding errors from gzip (#174, #176)
• truncated long error messages in the plain-text tokenizer (#172, #176)

Other

• bump aiohttp from 3.10.2 to 3.10.11 (#173, #176)

0.17.0

Features

• supported multi-modal embeddings from Azure AI Vision service (#162)

Fixes

• fixed incorrect calculation of image tokens for GTP-4o mini models (#166)
• bugfix: image tokenizer for chat completion requests ignored messages[*].content[*].image_url.detail field (#167, #169)

Other

• bump epam/ai-dial-ci from 1.9.3 to 1.9.4 (#163)

0.16.0

Features

• supported content parts for multi-modal models (GPT-4V and GPT-4o) (#154)
• added Docker healthcheck (#147)

Other

• bump epam/ai-dial-ci from 1.9.1 to 1.9.2 (#159)
• bump epam/ai-dial-ci from 1.9.2 to 1.9.3 (#161)
• bump starlette from 0.36.3 to 0.40.0; fastapi from 0.109.2 to 0.115.2 (#160)

0.15.0

Features

• supported message truncation for multi-modal models (GPT4V and GPT-4o) (#150, #153)
• supported streaming emulation for deployments specified in NON_STREAMING_DEPLOYMENTS env var (#157)
• introduced ELIMINATE_EMPTY_CHOICES env var as a replacement for FIX_STREAMING_ISSUES_IN_NEW_API_VERSIONS (#139)

Fixes

• fixed incorrect handling of usage reported by the upstream OpenAI model (#151)

Other

• bumped github workflow versions (#158)

0.14.0

Features

• added code=truncate_prompt_error to errors raised by prompt truncation algorithm (#142)

Fixes

• token usage isn't computed if upstream model has itself reported it (#138)
• changed type of created field in chat completion chunk from str to int (#141)
• improved display of DIAL links leading to the public bucket (#148)

Other

• bump aidial-sdk from 0.8.0 to 0.13.0 (#142)
• bump aiohttp from 3.9.4 to 3.10.2 (#143)
• bump cryptography from 42.0.5 to 43.0.1 (#146)

0.13.2

Fixes

• disabled retries in openai client accidentally introduced in 0.13.0 (#144) (#145)

0.13.1

Other

• bump setuptools from 68.2.2 to 70.0.0 (#134)
• bump epam/ai-dial-ci from 1.8.3 to 1.9.0 (#136)

0.13.0

Features

• supported completions (legacy) endpoints (#124, #127, #126, #129)
• supported deployment ids with colons (#125)

Fixes

• failing with user-facing error in case of an incorrect GPT-4 Vision request (#131)
• fixed content violation error handling for Dall-E-3 (#128)
• fixed token renewal logic (#120)

Other

• migrated from openai=0.28.1 to openai=1.33.0 (#105, #121)
• bump certifi from 2023.7.22 to 2024.7.4 (#132)
• bump epam/ai-dial-ci from 1.8.1 to 1.8.3 (#130)
• bump urllib3 from 2.0.7 to 2.2.2 (#119)
• bump zipp from 3.18.1 to 3.19.1 (#133)
• changed ownership (#123)

0.12.0

Features

• add display_message support for user-facing errors (#111)
• supported gpt4o models (#107)
  GPT-4o models should be declared as a comma-separated list via GPT4O_DEPLOYMENTS env vars

Other

• bump azure-identity from 1.15.0 to 1.16.1 (#115)
• bump epam/ai-dial-ci from 1.8.0 to 1.8.1 (#117)
• bump msal from 1.27.0 to 1.28.0 (#116)
• bump requests from 2.31.0 to 2.32.0
• bump tqdm from 4.66.1 to 4.66.3 (#102)