dev-python/jiter::bentoo
- Ebuilds: 1, Testing: 0.16.0 Description:
jiter is a fast, iterable JSON parser written in Rust and exposed to
Python through pyo3. Unlike the stdlib json module, it can parse a
document incrementally and offers configurable handling of floats
(float, Decimal or a lossless representation), duplicate keys and
partial/incomplete input, which makes it suitable for consuming
streaming JSON such as LLM responses. It is the JSON backend used by
pydantic-core and the OpenAI Python client.
Homepage:
https://github.com/pydantic/jiter/
https://pypi.org/project/jiter/
License: MIT Apache-2.0 Apache-2.0-with-LLVM-exceptions MIT MPL-2.0 Unicode-3.0
dev-python/openai::bentoo
- Ebuilds: 1, Testing: 3.3.1 Description:
The official OpenAI Python library. It provides synchronous and
asynchronous clients for the OpenAI REST API, covering responses, chat
completions, embeddings, images, audio, files, batches, fine-tuning,
vector stores and the realtime API, with request/response models typed
via pydantic. It also ships an Azure OpenAI client and, optionally, an
Amazon Bedrock provider.
Homepage:
https://github.com/openai/openai-python
https://pypi.org/project/openai/
License: Apache-2.0
dev-python/tavily-python::bentoo
- Ebuilds: 1, Testing: 0.7.27 Description:
Official Python client for the Tavily API, a web search engine built for
LLM and RAG pipelines. It exposes a synchronous TavilyClient and an
asyncio AsyncTavilyClient covering search, extract, crawl and map, plus
a TavilyHybridClient that combines Tavily results with a local vector
store. Results can be trimmed to a token budget, which is why tiktoken
is a hard runtime dependency. The PyPI distribution is named
tavily-python; the importable module is tavily.
Homepage:
https://github.com/tavily-ai/tavily-python
https://pypi.org/project/tavily-python/
License: MIT
dev-python/tiktoken::bentoo
- Ebuilds: 1, Testing: 0.14.0 Description:
tiktoken is a byte pair encoding (BPE) tokeniser from OpenAI. It splits
text into the integer tokens that GPT-family models actually consume, and
decodes them back, which is what token counting, context-window budgeting
and prompt truncation are built on. The hot loop is a Rust extension
module bound through pyo3, so it is roughly an order of magnitude faster
than pure-Python tokenisers. Encoding tables (r50k_base, p50k_base,
cl100k_base, o200k_base, o200k_harmony) are downloaded on first use and
cached locally.
Homepage:
https://github.com/openai/tiktoken/
https://pypi.org/project/tiktoken/
License: MIT Apache-2.0-with-LLVM-exceptions MIT Unicode-3.0