Add llm-prompt-default-max-tokens, open AI token limit fixes, parallel tool use fixes

What's Changed

• Fix breakage with Open AI's llm-chat-token-limit by @ahyatt in #77
• Fix Vertex and Open AI's parallel call tool use by @ahyatt in #78
• Add variable llm-prompt-default-max-tokens by @ahyatt in #79
• Fix how we look for ollama models in integration tests by @ahyatt in #80

Full Changelog: 0.17.3...0.17.4

More efficient, streaming text insertion, function calling conversational fixes

What's Changed

• Make streaming not repeatedly insert the same thing by @ultronozm in #72
• Fix error with ollama function results by @ultronozm in #74
• Fix bug involving multiple function calls with Claude by @ultronozm in #73
• Remove the debug logs output on streaming, to reduce log volume by @ahyatt in #75

Full Changelog: 0.17.2...0.17.3

New default models, Open AI temperature bugfix, function calling bugfixes

What's Changed

• Fix temperature calculation for llm-openai by @ultronozm in #61
• Fix bug in how we calculate Open AI number of calls in streaming by @ahyatt in #64
• Upgrade the default chat models for Claude and Open AI by @ahyatt in #65
• Only collect streaming function data when we have function calls by @ahyatt in #66
• Add integration testing by @ahyatt in #67
• Fix function calling conversations for Claude by @ahyatt in #68

New Contributors

• @ultronozm made their first contribution in #61

Full Changelog: 0.17.1...0.17.2

Function calling for ollama, ollama model function specificity, prompt bugfix

What's Changed

• fix typo by @conao3 in #52
• Make sure every model, even unknown ones, have some context size by @ahyatt in #54
• Add token count for llama3.1 model by @ahyatt in #55
• Add function calling for ollama by @ahyatt in #56
• Add model specifity for embeddings capabilities for ollama by @ahyatt in #57
• Test properly for prompt functions, including compiled functions by @ahyatt in #58
• Use the new /embed API for ollama by @ahyatt in #59

New Contributors

• @conao3 made their first contribution in #52

Full Changelog: 0.17.0...0.17.1

llm-prompt added, a way to define and fill prompts for LLMs

• Introduced llm-prompt for prompt management and creation from generators.
• Removed Gemini and Vertex token counting, because llm-prompt uses token
  counting often and it's best to have a quick estimate than a more expensive
  more accurate count.

Fix Open AI's GPT-4o context length

Open AI's GPT-4o has a context length for normal customers, and a higher one for corporate customers. We were using the latter, but now we'll use the normal one, which is most common.

Support for proxies

• Add support for HTTP/HTTPS proxies by @r0man in #49

Non-standard parameters and default-parameters

With this release, clients can set parameters that are not guaranteed to work with any model, but can be used to set specific parameters that could be unique to a model, when the client knows a specific model will be used. See llm-make-chat-prompt for details.

Clients can also set default chat parameters on models, which will apply to every chat with that model, unless overridden by the prompt parameters.

Migration to plz for requests, model context size updates for OpenAI and Gemini

• Move to plz backend, which uses curl. This helps move this package to a stronger foundation backed by parsing to spec. Thanks to @r0man for contributing the plz extensions that enable this, which are currently bundled in this package but will eventually become their own separate package.
• Add model context information for Open AI's GPT 4-o.
• Add model context information for Gemini's 1.5 models.

Fix for Vertex, getting GNU ELPA unstuck

0.14.2

Set version to 0.14.2