Releases: langroid/langroid
0.1.227
Minor updates:
- update to openai 1.23.2, which affects OpenAI Assistant API slightly
- allow DocChatAgentConfig to have vecdb = None, in case users want to set it later
0.1.226
DocumentParser: Improve plain text detection by taking into account continuation bytes.
This handles scenarios where multiple bytes can represent a character, e.g. non-Latin scripts such as Chinese, Arabic, etc.
0.1.225
TableChatAgent.run_code: use pd.eval instead of eval.
The former only accepts restricted data-frame related expressions, and is safer than eval.
0.1.224
DocChatAgent.ingest_docs: add window ids when not splitting.
This allows ingesting docs that have been already split via another library.
0.1.222
OpenAIGPT retry params changed: max retries 10 -> 5, exponential 2 -> 1.3
0.1.221
UnstructuredDocParser: fix partition_doc arg to allow bytes
0.1.219
DocChatAgent: allow ingest_doc_paths to take bytes not just doc paths
0.1.218
JSON parsing (relevant for parsing tools/function calls): improved repair of un-quoted keys or values via yaml.
Intended to handle cases where a weak LLM may produce a JSON-like string
containing, e.g. "rent": DO-NOT-KNOW, where it "forgot" to put quotes on the value,
or city: "New York" where it "forgot" to put quotes on the key.
It will even handle cases like 'address: do not know'.
0.1.217
Further improvements in batch/concurrent running of DocChatAgent.
See test_doc_chat_batch
here https://github.com/langroid/langroid/blob/main/tests/main/test_doc_chat_agent.py
Now to run multiple instances of DocChatAgent
concurrently, you can use either run_batch_tasks
where you pass in a single agent,
or run_batch_task_gen
where you pass in a task-generator.
0.1.216
- Concurrency: Improvements to allow concurrent runs of
DocChatAgent
, see the
test_doc_chat_agent.py
in particular thetest_doc_chat_batch()
; - Batch running of tasks: New task run utility:
run_batch_task_gen
where a task generator can be specified, to generate one task per input.