Skip to content

Rework capabilities pages with LFM3 and LFM2/LFM2.5 tabs - #179

Open
iamleonie wants to merge 17 commits into
mainfrom
leonie/capabilities-by-task
Open

Rework capabilities pages with LFM3 and LFM2/LFM2.5 tabs#179
iamleonie wants to merge 17 commits into
mainfrom
leonie/capabilities-by-task

Conversation

@iamleonie

@iamleonie iamleonie commented Nov 13, 2026 •
edited by alay2shah
Loading

Copy link
Copy Markdown
Contributor

Summary

  • Builds on Add Thinking key-concepts page #174, now merged into main. Updated against main while preserving the reviewed thinking-off wording, optional LFM3-Audio reasoning note, model-specific reasoning retention, vision serving setup, and Docs audit: discovery, assistant guidance, and LFM3 recommendations #172 migration guidance.
  • Add a Speech Generation page; remove Omni Capabilities (redirected to Audio Capabilities) and fold its examples into Audio, Speech Generation, Vision, and Tool Use.
  • Add LFM3 and LFM2/LFM2.5 tabs (Audio and Omni tabs on the audio pages) to Chat Template, Thinking, Tool Use, and Vision. Tabs share wording and examples and differ only where the code differs.
  • Keep the prose model-agnostic: model-specific details such as sampling values, languages, and frame rates point to the model pages.
  • Drop the shared generate() helpers; every example shows its full code.
  • Keep examples and headings close to the original pages.
  • Refresh all example outputs and result images from Colab runs of the cookbook notebooks (Align capability guide notebooks with the docs and add an LFM3-VL guide cookbook#150), which use the same code.
  • Point model pages to the new pages.

Validation of main update

  • Mintlify build validation passed.
  • URL snapshot check passed: 235 active URLs.
  • Both GitHub checks passed; all 86 Python code blocks parsed successfully.
  • Deployed preview verified for Thinking, Chat Template, Audio Capabilities, Speech Generation, Vision Capabilities, Tool Use, and Migration Guide. Internal links to these pages resolve to existing section anchors.
  • Content review identified follow-up corrections for direct-answer parsing, Omni setup dependencies, text-only output controls, and the spoken tool-call example. Those corrections are pending review and are not included in this main update.
  • Model inference was not rerun. This update does not approve the PR for merge.

@iamleonie
Add Thinking key-concepts page
aa74f5c 
Explains how to use thinking models, separate reasoning from the answer,
switch thinking off where supported, and which models think.
@iamleonie
Document thinking switches in the chat template page
aea3991 
Covers enable_thinking and preserve_thinking for LFM3, and how LFM2.5
thinking-model templates handle reasoning. Fixes the LFM2/LFM2.5 anchor.
@iamleonie
Standardize Reasoning sections on thinking-model pages
db3809c 
Use one template on every thinking-model page and link to the Thinking
page.
@iamleonie
Say LFM3-1.7B can switch thinking off but it isn't recommended
82018fd
@iamleonie
Fix grammar in thinking FAQ
1961eca
@iamleonie
Add LFM3 and LFM2.5 snippets and replace Omni page in navigation
e27e3eb 
Add chat prompt and tool call snippets for LFM3 and LFM2.5, rename the
tool call token snippet, list Speech Generation instead of Omni
Capabilities in the sidebar, and redirect the Omni Capabilities URL.
@iamleonie
Point model pages to the new capabilities pages
0ff27af 
Replace Omni Capabilities links and cards with Audio Capabilities,
Speech Generation, Vision Capabilities, and Tool Calling, and link
video examples to the Video understanding section.
@iamleonie
Add LFM2/LFM2.5 tab to the Thinking text example
5ddd3a6 
Show the text example for LFM3 and LFM2/LFM2.5 thinking models in tabs,
and make the switch-off, model-choice, and FAQ prose model-agnostic.
@iamleonie
Restructure the Tool Calling page by input type
4476012 
Show LFM3 and LFM2/LFM2.5 code in tabs, give text, vision, and audio and
omni models their own sections with full examples, flatten the defer
loading section, and make the prose model-agnostic.
@iamleonie
Split speech output into a Speech Generation page and merge Omni into…
9bbcea6 
… Audio

Add the Speech Generation page for text-to-speech, spoken chat, and spoken
answers. Restructure Audio Capabilities around one shared setup with
LFM3-Audio and LFM3-Omni tabs, fold the Omni Capabilities examples into
it, and remove the Omni Capabilities page.
@iamleonie
Restructure Vision Capabilities with LFM3 and LFM2/LFM2.5 tabs
921d6a3 
Show each example once with tabs where the code differs, add the omni
voice and video soundtrack examples, and refresh outputs and result
images from the cookbook notebook runs.
@iamleonie
Restructure Chat Template with LFM3 and LFM2/LFM2.5 tabs
010c69e 
Use the same messages for both families, list special tokens in tables
with the original wording, and add audio and omni examples.
@iamleonie
Refresh Tool Use outputs from the cookbook notebook runs
e3f97cf
@iamleonie
iamleonie requested a review from mlabonne as a code owner November 13, 2026 16:46
@mintlify

mintlify Bot commented Nov 13, 2026 •
edited
Loading

Copy link
Copy Markdown

Preview deployment for your docs. Learn more about Mintlify Previews.

Project Status Preview Updated
liquid-docs 🟢 Ready View Preview Nov 14, 2026, 3:37 PM

💡 Tip: Enable Automations to automatically generate PRs for you.

@iamleonie
iamleonie requested review from alay2shah and removed request for mlabonne November 13, 2026 16:47
@iamleonie
iamleonie changed the base branch from main to leonie/thinking-page November 13, 2026 16:50
@mlabonne

mlabonne commented Nov 13, 2026

Copy link
Copy Markdown
Collaborator

We're missing LFM3 in the migration guide, can you check if everything is up-to-date with LFM3?

@alay2shah alay2shah mentioned this pull request Nov 13, 2026
Merged
alay2shah added a commit that referenced this pull request Nov 13, 2026
@iamleonie @alay2shah
Fix docs discovery, assistant guidance, and LFM3 recommendations (#172)
a0eaf0b 
## Summary

Model pages outside the navigation were missing from search and
assistant results, and general recommendations still led with older
models. This audit makes model documentation discoverable and puts LFM3
first in general guidance while retaining useful older models.

- **Search and indexing:** enable indexing for all pages; retain
deprecated model and tool references with warnings. Exclude
docs-maintainer files through `.mintignore` and mark the contributing
page `noindex`.
- **Assistant:** remove the CSS rule that hid the legacy assistant
entry. Add instructions to cite measured performance, use Pipette,
follow exact model-page format links for verified download sizes, and
use published evaluation sources without inventing repositories or
scores.
- **LFM3 guidance:** update the FAQ, use-case evaluation, migration
guide, and troubleshooting. Compare LFM3, LFM2.5, and LFM2; lead with
LFM3 text, vision, audio, and omni models; retain smaller, specialized,
and larger MoE alternatives with explicit reasons.
- **Version-specific instructions:** distinguish LFM3 XML tool calls and
`qwen3_coder` parsing from LFM2/LFM2.5 formats and `lfm2` parsing. Scope
chat templates, LoRA module examples, generation settings, and context
limits to the selected model.
- **Deprecations and examples:** point LFM2-2.6B, LFM2-2.6B-Exp, and
LFM2-Audio directly to current replacements in notices, metadata, and
the deprecations table. Add missing notices to four audio examples, a
newer-model note to the invoice extractor, and warnings to archived
bundling references.
- **Formats and quickstarts:** expose the verified LFM3-Audio and
LFM3-Omni ONNX repositories on their model pages and in the matrix.
Organize ColBERT quickstarts into Retrieval and Reranking, with
library/runtime choices inside Reranking; match the embedding page's
“Sentence Transformers” label.
- **Other audit fixes:** add eight legacy/mistyped URL redirects,
improve embeddings/reranking/RAG terminology, make internal example
links root-relative, and correct the Ollama model-name typo.

## Validation

- Mintlify build validation passed.
- GitHub Check docs and Check link snapshot passed on `1b21fbe`.
- Updated preview content verified for the FAQ, evaluation and migration
guides, troubleshooting, deprecation notice, and ONNX model
buttons/matrix.
- Link snapshot: 233 active URLs verified.
- All 89 distinct internal links and snippet imports in the latest 11
changed files resolve; 20 Python snippets parse.
- New Audio and Omni ONNX repositories verified against their official
Hugging Face pages.
- Updated deprecation notices resolve directly to models that are not
deprecated.

Removing the assistant CSS restriction does not establish that Mintlify
has enabled chat permissions; project configuration still controls
assistant availability.

## Preview review

- [FAQ](https://liquidai-docs-audit.mintlify.site/lfm/help/faqs)
- [Use Case
Evaluation](https://liquidai-docs-audit.mintlify.site/guides/use-case-evaluation)
- [Migration
Guide](https://liquidai-docs-audit.mintlify.site/guides/migration-guide)
-
[Troubleshooting](https://liquidai-docs-audit.mintlify.site/lfm/help/troubleshooting)
-
[Deprecations](https://liquidai-docs-audit.mintlify.site/lfm/help/deprecations)
- [Model
Matrix](https://liquidai-docs-audit.mintlify.site/lfm/models/complete-library#complete-model-matrix)
- [ColBERT
Quickstart](https://liquidai-docs-audit.mintlify.site/lfm/models/lfm25-colbert-350m#quick-start)

## Related work

- #174 and its stacked PR #179 cover Thinking and capability-page
organization.
- #175 covers the Android SDK-to-llama.cpp migration.
- Runtime quickstart rewrites and training/inference notebook
modernization remain follow-up work.

A three-way check against #179 found two small conflicts in
`guides/migration-guide.mdx`: the shared tool-call snippet import and
component invocation. When integrating #179, preserve this PR's LFM3
guidance, adopt its `LFM25ToolCallExample` name with paired JSX tags,
and keep the unused `ChatMLExample` import removed.

---------

Co-authored-by: Leonie Monigatti <iamleonie@users.noreply.github.com>
Co-authored-by: alay2shah <alay0shah@gmail.com>
alay2shah and others added 3 commits November 13, 2026 16:06
Base automatically changed from leonie/thinking-page to main November 14, 2026 15:35

This branch was successfully deployed

1 active deployment
staging — 55c10316 Deployed Nov 14, 2026 by mintlify[bot]
View deployment
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

@alay2shah alay2shah Awaiting requested review from alay2shah

At least 1 approving review is required to merge this pull request.

Assignees

No one assigned

Labels

None yet

Projects

None yet

Milestone

No milestone

Development

Successfully merging this pull request may close these issues.

3 participants

@iamleonie @mlabonne @alay2shah