Skip to content

fix model-id error caused by previous refactor - #46

Merged
jakelorocco merged 1 commit into
mainfrom
jal/fix-formatter-model-id-bug
Aug 15, 2025
Merged

jakelorocco merged 1 commit into
mainfrom
jal/fix-formatter-model-id-bug

Conversation

@jakelorocco

@jakelorocco jakelorocco commented Aug 13, 2025 •

Copy link
Copy Markdown
Contributor

fixes: #37

in a previous version of the code, the model_id was always being stringified here before being checked; this is not the case now, which would cause the check to incorrectly pass even though the value was None

all formatter tests still passed, and I added two new tests to check for similar situations

@jakelorocco
jakelorocco merged commit 77d2f08 into main Aug 15, 2025
@jakelorocco
jakelorocco deleted the jal/fix-formatter-model-id-bug branch August 15, 2025 08:25
akihikokuroda pushed a commit to akihikokuroda/mellea that referenced this pull request May 27, 2026
spzala pushed a commit to spzala/mellea that referenced this pull request Oct 6, 2026
… io.yaml (generative-computing#1706)

* chore(adapters): bump granitelib-core pin for fixed requirement-check io.yaml

ibm-granite/granitelib-core-r1.0 PR generative-computing#46 (merged 2026-10-05, commit
c4b4fc6ec4fa05b6c8f02b25ac8822da085962b5) drops the colon from the
requirement-check aLoRA instruction (`<requirements>: {requirement}` ->
`<requirements> {requirement}`), so the invocation sequence now survives
tokenisation and the adapter activates as published.

The pin covers all three core adapter functions (context-attribution,
requirement-check, uncertainty); only requirement-check's files changed
between the old and new revisions.

Removes the strict xfail from
test_requirement_check_adapter_moves_score, which needs a GPU and is
skipped locally.

Closes generative-computing#1699, generative-computing#1679.

Assisted-by: Claude Code
Signed-off-by: Nigel Jones <jonesn@uk.ibm.com>

* docs: sync stale granitelib-core SHA references to the new pin

Two mechanical review comments on PR generative-computing#1706 (code-review panel): the
schema-migrations tutorial's sample output and the uncertainty Ollama
build script still showed the old d0a2a96a pin after the catalog bump.

Assisted-by: Claude Code
Signed-off-by: Nigel Jones <jonesn@uk.ibm.com>

---------

Signed-off-by: Nigel Jones <jonesn@uk.ibm.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

Error in template lookup logic

1 participant