Problemsby Ibis

Data and AI struggles

Serving models on MLflow is harder than it should be

MLOps9 posts from 8 people6d active+00 posts in the last 7 days, 3 the 7 days before (fading)

Posts per day9 posts, Sep 9 to Oct 21

The posts behind it

9, newest first

GitHub issuesOct 11 [BUG] Tensor-based signature silently casts DataFrame input instead of validating it, corrupting values...`_reshape_and_cast_pandas_column_values` (`mlflow/models/utils.py`), all three branches fail to validate: The `dtype=tensor_spec.type` cast runs **before** `_enforce_tensor_spec`, whose dtype guard is `if clean_tensor_type(actual_type) != expected_type: raise ...` - by then...mlflow/mlflow · hdp014 repliesOpen ↗GitHub issuesOct 10 [BUG] rename_registered_model drops registered model tags in the file store...(e.g., Linux Ubuntu 16.04)**: macOS 26.6 - **Python version**: 3.10.17 ### Describe the problem With the file-based model registry, `rename_registered_model` drops the registered model's tags. Expected `{'k': 'v'}` after the rename, got `{}`. The SQLAlchemy store keeps them. ###...mlflow/mlflow · harupy1 repliesOpen ↗GitHub issuesOct 10 [BUG] @mlflow.trace records self in span inputs> [!WARNING] > Before submitting a PR, please make sure that: > - A maintainer has triaged this issue and applied the `ready` label > - This issue has no assignee > - No duplicate PR exists > > PRs not meeting these requirements may be automatically closed. ### Issues Policy...mlflow/mlflow · harupy1 repliesOpen ↗GitHub issuesOct 6 [BUG] dump_input_data fails for every dict input (iterates keys instead of items)> [!WARNING] > Before submitting a PR, please make sure that: > - A maintainer has triaged this issue and applied the `ready` label > - This issue has no assignee > - No duplicate PR exists > > PRs not meeting these requirements may be automatically closed. ### Issues Policy...mlflow/mlflow · maharanay221 reactions, 2 repliesOpen ↗GitHub issuesOct 6 [BUG] `search_registered_models` / `search_model_versions` raise ValueError for `IS NULL` / `IS NOT NULL` filters> [!WARNING] > Before submitting a PR, please make sure that: > - A maintainer has triaged this issue and applied the `ready` label > - This issue has no assignee > - No duplicate PR exists > > PRs not meeting these requirements may be automatically closed. ### Issues Policy...mlflow/mlflow · ParamTanna3 repliesOpen ↗GitHub issuesOct 6 [FR] Graceful Handling of Unavailable Multipart Uploads...### Proposal Summary In https://github.com/mlflow/mlflow/issues/25974 I described an issue with multipart uploads on Azure. The issue has been fixed in the meantime, but a certain limitation remains: When the client tries to initiate the multipart upload with insufficient...mlflow/mlflow · peoe1 repliesOpen ↗GitHub issuesOct 5 [BUG] Tensor schema enforcement fails on non-default DataFrame row indexes...3.13.3. - pandas: 3.0.6. - NumPy: 2.5.3. ### Describe the problem `PyFuncModel.predict` fails to enforce a tensor signature when a DataFrame has row labels such as `[10, 20]`. The same values with a default index predict correctly. Filtering a DataFrame can leave these row...mlflow/mlflow · aswanth-072 repliesOpen ↗GitHub issuesOct 3 [FR] Configurable API base for the TypeSafe gateway provider...could send System One calls to it without code changes. > #### Why is it currently difficult to achieve this use case? `TypeSafeProvider.base_url` returns the `TYPESAFE_API_BASE_URL` constant and `TypeSafeConfig` only has `typesafe_api_key`, so the only way to change the host is...mlflow/mlflow · SamratDuttaOfficial2 repliesOpen ↗GitHub issuesSep 9 Vulnerability - [BUG]> [!WARNING] > Before submitting a PR, please make sure that: > - A maintainer has triaged this issue and applied the `ready` label > - This issue has no assignee > - No duplicate PR exists > > PRs not meeting these requirements may be automatically closed. ### Issues Policy...mlflow/mlflow · kiran71319-beep8 repliesOpen ↗

Companies and products named

MLflow Named8 posts

GitHub Named2 posts

pandas Named1 post

SageMaker Named1 post

SQLite Named1 post

Microsoft Azure

About this problem

Evidence

9 posts from 8 people in 1 place, about 2 a week over 33 days. Mostly on mlflow/mlflow. Tools named alongside: GitHub, pandas, SageMaker, SQLite.

Frustration Frustration 0 of 3· Seen on GitHub issues

How it was grouped

Posts that state a pain and match the "model-serving" rule about MLflow. First post Sep 9, 2026, latest Oct 11, 2026. Corroborated: 3 or more posts from 2 or more people or places. Method

The same problem elsewhere

Elsewhere15 postsDeploying and serving models is harder than training Elsewhere4 postsServing models on Ollama is harder than it should be Elsewhere2 postsServing models on ChatGPT is harder than it should be Elsewhere2 postsServing models on OpenAI is harder than it should be

Other MLflow problems

Same product22 postsMLflow users keep filing bugs Same product7 postsMLflow upgrades break working setups Same product5 postsTrouble with MLflow list Same product4 postsTrouble with MLflow metric Same product3 postsTrouble with MLflow experiments Same product3 postsMLflow rate limits and usage caps block work Same product2 postsMLflow backfills are slow and risky Same product2 postsTrouble with MLflow page

History

MomentumOct 17, 2026steady to fading

MomentumOct 16, 2026fading to steady

MomentumOct 13, 2026steady to fading

MomentumOct 12, 2026rising to steady

AddedOct 8, 2026Serving models on MLflow is harder than it should be (6 posts)