Liquid Foundation Model 2 support by winglian · Pull Request #2905 · axolotl-ai-cloud/axolotl

winglian · 2025-07-11T14:59:01Z

Description

LFM2 blog post: https://www.liquid.ai/blog/liquid-foundation-models-v2-our-second-series-of-generative-ai-models

added example with packing (seems to work correctly even though causal_conv1d support isn't quite correct for packed sequences yet, see #state-spaces/mamba/244)

to add proper packing implemented in modeling code might look like this
https://github.com/zigzagcai/varlen_mamba/blob/feat/add-cu_seqlens/mamba_ssm/modules/mamba_simple.py#L188-L198

if cu_seqlens is not None:
    # naive pure python implementation of varlen causal_conv1d
    for i, s in enumerate(cu_seqlens[1:-1]):
        x = torch.cat((x[..., :s + i*(self.d_conv - 1)], torch.zeros_like(x[..., :(self.d_conv - 1)]), x[..., s + i*(self.d_conv - 1):]), dim=2)
    mask = torch.cat([torch.cat((torch.full((s,), True, dtype=torch.bool, device=x.device), 
                                 torch.full((self.d_conv - 1,), False, dtype=torch.bool, device=x.device)), dim=0) 
                      for s in (cu_seqlens[1:] - cu_seqlens[:-1])], dim=0)
    x = self.act(self.conv1d(x)[:, :, mask])
else:
    x = self.act(self.conv1d(x)[..., :seqlen])

Summary by CodeRabbit

Documentation
- Added a README file for the Liquid Foundation Models 2 (LFM2) example, including installation instructions for the latest transformers library.
New Features
- Introduced a configuration file for training the LFM2-350M model, detailing dataset settings and training parameters.

coderabbitai · 2025-07-11T14:59:08Z

Walkthrough

Two new files have been added under the LFM2 examples: a README providing installation guidance for the latest transformers library, and a YAML configuration file detailing dataset, training, and logging parameters for the LFM2-350M model.

Changes

File(s)	Change Summary
examples/lfm2/README.md	Added a README with installation instructions for using LFM2 with the latest transformers branch.
examples/lfm2/lfm2-350m-fft.yaml	Added a YAML config specifying dataset, training, and logging parameters for LFM2-350M training.

Suggested reviewers

NanoCode012

Poem

In the warren of code, a new path appears,
With configs and docs, the intent is clear.
LFM2 now trains, with YAML in tow,
And README instructions for those in the know.
A hop and a skip, let experiments begin—
May your models be swift and your losses be thin! 🐇✨

📜 Recent review details

Configuration used: CodeRabbit UI
Review profile: CHILL
Plan: Pro

📥 Commits

Reviewing files that changed from the base of the PR and between 310caf2 and 7c6c10e.

📒 Files selected for processing (1)

examples/lfm2/lfm2-350m-fft.yaml (1 hunks)

🚧 Files skipped from review as they are similar to previous changes (1)

examples/lfm2/lfm2-350m-fft.yaml

⏰ Context from checks skipped due to timeout of 90000ms. You can increase the timeout in your CodeRabbit configuration to a maximum of 15 minutes (900000ms). (2)

GitHub Check: pre-commit
GitHub Check: preview

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

🪧 Tips

Chat

There are 3 ways to chat with CodeRabbit:

Review comments: Directly reply to a review comment made by CodeRabbit. Example:
- I pushed a fix in commit <commit_id>, please review it.
- Explain this complex logic.
- Open a follow-up GitHub issue for this discussion.
Files and specific lines of code (under the "Files changed" tab): Tag @coderabbitai in a new review comment at the desired location with your query. Examples:
- @coderabbitai explain this code block.
- @coderabbitai modularize this function.
PR comments: Tag @coderabbitai in a new PR comment to ask questions about the PR branch. For the best results, please provide a very specific query, as very limited context is provided in this mode. Examples:
- @coderabbitai gather interesting stats about this repository and render them as a table. Additionally, render a pie chart showing the language distribution in the codebase.
- @coderabbitai read src/utils.ts and explain its main purpose.
- @coderabbitai read the files in the src/scheduler package and generate a class diagram using mermaid and a README in the markdown format.
- @coderabbitai help me debug CodeRabbit configuration file.

Support

Need help? Create a ticket on our support page for assistance with any issues or questions.

Note: Be mindful of the bot's finite context window. It's strongly recommended to break down tasks such as reading entire modules into smaller chunks. For a focused discussion, use review comments to chat about specific files and their changes, instead of using the PR comments.

CodeRabbit Commands (Invoked using PR comments)

@coderabbitai pause to pause the reviews on a PR.
@coderabbitai resume to resume the paused reviews.
@coderabbitai review to trigger an incremental review. This is useful when automatic reviews are disabled for the repository.
@coderabbitai full review to do a full review from scratch and review all the files again.
@coderabbitai summary to regenerate the summary of the PR.
@coderabbitai generate docstrings to generate docstrings for this PR.
@coderabbitai generate sequence diagram to generate a sequence diagram of the changes in this PR.
@coderabbitai resolve resolve all the CodeRabbit review comments.
@coderabbitai configuration to show the current CodeRabbit configuration for the repository.
@coderabbitai help to get help.

Other keywords and placeholders

Add @coderabbitai ignore anywhere in the PR description to prevent this PR from being reviewed.
Add @coderabbitai summary to generate the high-level summary at a specific location in the PR description.
Add @coderabbitai anywhere in the PR title to generate the title automatically.

CodeRabbit Configuration File (`.coderabbit.yaml`)

You can programmatically configure CodeRabbit by adding a .coderabbit.yaml file to the root of your repository.
Please see the configuration documentation for more information.
If your editor has YAML language server enabled, you can add the path at the top of this file to enable auto-completion and validation: # yaml-language-server: $schema=https://coderabbit.ai/integrations/schema.v2.json

Documentation and Community

Visit our Documentation for detailed information on how to use CodeRabbit.
Join our Discord Community to get help, request features, and share feedback.
Follow us on X/Twitter for updates and announcements.

github-actions · 2025-07-11T15:04:18Z

📖 Documentation Preview: https://6871e07b86558bdf0f609fce--resonant-treacle-0fd729.netlify.app

Deployed on Netlify from commit 7c6c10e

coderabbitai

Actionable comments posted: 2

🧹 Nitpick comments (3)

examples/lfm2/lfm2-350m-fft.yaml (2)
21-26: Blank WandB fields may cause silent auto-init.

When wandb_project is empty, WandB will fall back to your global default or open an anonymous run, which can be confusing.

Either:
-wandb_project:
+wandb_project: lfm2-experiments
or explicitly disable WandB with wandb: disabled in the launcher.

17-20: pad_to_sequence_len: false + long context may hurt throughput.

Without sample packing and without padding, each example is individually 0-padded to its own length ⇒ poor GPU utilisation on chat datasets with large length variance.

Consider enabling padding to fixed sequence_len once sample-packing support lands, or set sequence_len closer to your dataset’s 90-th percentile length.
examples/lfm2/README.md (1)
3-7: Pin the transformers commit for reproducibility.

pip install git+https://github.com/huggingface/transformers.git pulls moving main, which breaks reproducibility and may introduce incompatible API changes.
pip install git+https://github.com/huggingface/transformers.git@<commit-sha>
Replace <commit-sha> with the commit you validated this example against and mention it in the README.

📜 Review details

Configuration used: CodeRabbit UI
Review profile: CHILL
Plan: Pro

📥 Commits

Reviewing files that changed from the base of the PR and between 03b2a11 and 9ea9a7d.

📒 Files selected for processing (2)

examples/lfm2/README.md (1 hunks)
examples/lfm2/lfm2-350m-fft.yaml (1 hunks)

⏰ Context from checks skipped due to timeout of 90000ms. You can increase the timeout in your CodeRabbit configuration to a maximum of 15 minutes (900000ms). (2)

GitHub Check: pre-commit
GitHub Check: preview

🔇 Additional comments (1)

examples/lfm2/lfm2-350m-fft.yaml (1)

30-33: Learning rate looks aggressive for 350 M model.

5e-5 with AdamW-fused and cosine schedule tends to diverge when fine-tuning full-precision 350 M models on long-context data. Common practice is 1–2e-5.

Please sanity-check with a short ~200-step pilot run and inspect loss curves.

djsaunde

I need to read about this 👀

djsaunde · 2025-07-11T15:04:07Z

@@ -0,0 +1,7 @@
+# Liquid Foundation Models 2


TIL about liquid foundation models

salmanmohammadi · 2025-07-11T15:10:18Z

+LFM2 support in transformers exists in the main branch, but is not yet included in the transformers release.
+
+```bash
+pip install git+https://github.com/huggingface/transformers.git


--upgrade --no-deps maybe?

winglian added 2 commits July 11, 2025 10:09

LFM2 support

82be91f

docs

9ea9a7d

winglian requested review from djsaunde and salmanmohammadi July 11, 2025 14:59

coderabbitai Bot reviewed Jul 11, 2025

View reviewed changes

Comment thread examples/lfm2/lfm2-350m-fft.yaml

Comment thread examples/lfm2/lfm2-350m-fft.yaml

djsaunde approved these changes Jul 11, 2025

View reviewed changes

salmanmohammadi reviewed Jul 11, 2025

View reviewed changes

winglian added 2 commits July 11, 2025 12:44

packing seems to work

ec848ac

update install to force install in case already on dev version

310caf2

winglian added the ready to merge label Jul 11, 2025

default to use chunked cross entropy

7c6c10e

winglian merged commit 9a8073e into main Jul 12, 2025
4 checks passed

winglian deleted the lfm2 branch July 12, 2025 15:41

coderabbitai Bot mentioned this pull request Aug 14, 2025

Various fixes for VLMs #3063

Merged

winglian removed the ready to merge label Aug 18, 2025

coderabbitai Bot mentioned this pull request Oct 9, 2025

feat: add lfm2 family and latest moe model #3208

Merged

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

Liquid Foundation Model 2 support#2905

Liquid Foundation Model 2 support#2905
winglian merged 5 commits into
mainfrom
lfm2

winglian commented Jul 11, 2025 •

edited by coderabbitai Bot

Loading

Uh oh!

coderabbitai Bot commented Jul 11, 2025 •

edited

Loading

Chat

Support

CodeRabbit Commands (Invoked using PR comments)

Other keywords and placeholders

CodeRabbit Configuration File (`.coderabbit.yaml`)

Documentation and Community

Uh oh!

github-actions Bot commented Jul 11, 2025 •

edited

Loading

Uh oh!

coderabbitai Bot left a comment

Uh oh!

Uh oh!

Uh oh!

djsaunde left a comment

Uh oh!

djsaunde Jul 11, 2025

Uh oh!

salmanmohammadi Jul 11, 2025

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

Uh oh!

Conversation

winglian commented Jul 11, 2025 • edited by coderabbitai Bot Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Description

Summary by CodeRabbit

Summary by CodeRabbit

Uh oh!

coderabbitai Bot commented Jul 11, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Walkthrough

Changes

Suggested reviewers

Poem

Chat

Support

CodeRabbit Commands (Invoked using PR comments)

Other keywords and placeholders

CodeRabbit Configuration File (.coderabbit.yaml)

Documentation and Community

Uh oh!

github-actions Bot commented Jul 11, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

coderabbitai Bot left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

djsaunde left a comment

Choose a reason for hiding this comment

Uh oh!

djsaunde Jul 11, 2025

Choose a reason for hiding this comment

Uh oh!

salmanmohammadi Jul 11, 2025

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

winglian commented Jul 11, 2025 •

edited by coderabbitai Bot

Loading

coderabbitai Bot commented Jul 11, 2025 •

edited

Loading

CodeRabbit Configuration File (`.coderabbit.yaml`)

github-actions Bot commented Jul 11, 2025 •

edited

Loading