Experiments on Codex are hard to reproduce and track
MLOps2 posts from 2 people2d active+11 post in the last 7 days, 0 the 7 days before (steady)Unverified
Posts per day
The posts behind it
2, newest first[BUG] suppress_logging leaves logger levels changed after exit### Bug Description `suppress_logging()` does not restore a logger's configured level when its body raises. A failed ChromaDB search can therefore leave the HNSW logger at `ERROR`, hiding later warnings. Both `ChromaDBClient.search()` and `asearch()` use this helper around the...crewAIInc/crewAI · lakers-abingOpen ↗[Bug]: MarkdownNodeParser closes a code block on a shorter backtick fence...of the 22 failures reports missing NLTK `stopwords` data; the resource downloader is blocked by its network security check in this environment. I am not claiming a clean full-suite result. Related PR [#22123](https://github.com/run-llama/llama_index/pull/22123) is closed without...run-llama/llama_index · galleonliOpen ↗
Companies and products named
About this problem
Evidence
2 posts from 2 people in 2 places, about 1 a week over 19 days. Mostly on run-llama/llama_index, crewAIInc/crewAI. Tools named alongside: GitHub, PyPI, LlamaIndex, Chroma.
Frustration Frustration 0 of 3· Seen on GitHub issues
How it was grouped
Posts that state a pain and match the "ml-reproducibility" rule about Codex. First post Sep 28, 2026, latest Oct 16, 2026. Unverified: fewer than 3 posts, or one person or place only. Method
The same problem elsewhere
ML experiments are hard to reproduce and trackExperiments on ClickHouse are hard to reproduce and trackExperiments on LlamaIndex are hard to reproduce and trackExperiments on Apache Superset are hard to reproduce and trackExperiments on GitHub are hard to reproduce and trackExperiments on DuckDB are hard to reproduce and track
Other Codex problems
Codex fails to return valid structured outputTrouble with Codex localCodex users keep filing bugsTrouble with Codex windowsTrouble with Codex desktopCodex rate limits and usage caps block workCodex runs out of context on real workTrouble with Codex extensionTrouble with Codex selectionCodex lock-in makes switching costlyCodex outages and slowdowns disrupt workCodex error messages are cryptic
History
Experiments on Codex are hard to reproduce and track (2 posts)