Add Nanbeige4.2-3B export support - #4712
Open
micwill755 wants to merge 1 commit into
Open
micwill755 wants to merge 1 commit into
micwill755 wants to merge 1 commit into
Conversation
This branch has not been deployed
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Adds TensorRT Edge Exporter support for
Nanbeige/Nanbeige4.2-3B.Nanbeige uses a looped-transformer architecture with 22 physical decoder layers executed twice. This change:
EdgeSpec.trust_remote_codemodel.Motivation and context
Nanbeige4.2-3B reuses the same transformer blocks across multiple loops while requiring independent KV-cache state for each logical layer execution. Existing causal-LM export paths only process each physical layer once and therefore cannot represent this architecture correctly.
The model's remote code also targets an older Transformers version. With the current environment, its non-persistent RoPE frequency buffers may be zero-filled or invalid after loading. The compatibility helper deterministically reconstructs those buffers from the model configuration before eager parity capture.
Validation
Tested end-to-end with
Nanbeige/Nanbeige4.2-3Bin FP16:0.0084640.0468750.001699.4%15.948 ms9.446 ms1.688xTargeted Edge Exporter unit tests and pre-commit formatting/lint hooks pass locally.
Dependencies
Nanbeige/Nanbeige4.2-3B.Type of change
Checklist