[contrib] Add OLMoE-1B-7B-0924 NeuronX port

[dhwanw]

Description

NeuronX Distributed Inference port of allenai/OLMoE-1B-7B-0924, a 7B-parameter Mixture of Experts model with 1B active parameters per token. OLMoE uses 64 experts with top-8 routing, Q/K normalization (RMSNorm applied per-head after Q/K projection), and SwiGLU expert MLPs.

Model Information

Model Name: OLMoE-1B-7B-0924
Model Architecture: Decoder-only Mixture of Experts transformer -- 64 experts with top-8 routing, 16 MHA heads, 16 layers, Q/K RMSNorm, RoPE, SwiGLU MLP experts
Purpose: Text generation with sparse MoE architecture (1B active of 7B total parameters)

Checklist

Required Components

• Accuracy Test (test/integration/test_model.py)
  • Validates model generation and coherence
  • Performance benchmarks (TTFT, throughput)
  • Test can compile and run the model on Neuron
• README.md with the following sections:
  • Usage Example: Clear code example showing how to use the model
  • Compatibility Matrix: Table showing tested Neuron SDK versions and instance types
  • Example Checkpoints: Links to compatible model checkpoints
  • Testing Instructions: Command to run the test suite for the model
• Source Code (src/)
  • Modeling code following NxD Inference patterns

Optional Components

• Unit Tests (CPU or Neuron-based)

Folder Structure

/contrib/models/OLMoE-1B-7B-0924/
  README.md
  /src
    modeling_olmoe.py
  /test
    /integration
      test_model.py

Testing

Model was compiled and tested with TP=1, batch_size=1, seq_len=128, bfloat16 on trn1.32xlarge.

Test Results:

Test	Status	Result
Smoke Test	✅ PASS	Model loads successfully
Greedy Token Matching	✅ PASS	81.0% average (7/10 prompts at 100%)
Teacher-Forced Match	✅ PASS	98.0% average
Throughput	✅ PASS	2.9 tok/s

Compatibility

Tested with:

• Neuron SDK Version(s): 2.22
• Instance Type(s): trn1.32xlarge
• PyTorch Version: 2.9
• Python Version: 3.10
• Configuration: TP=1, batch_size=1, seq_len=128, bfloat16

Additional Information

• 64 experts with top-8 routing: Uses MoENeuronConfig for compilation. Expert routing is deterministic (softmax + top-k selection).
• Q/K normalization: Applies RMSNorm to Q and K projections after splitting heads, similar to OLMo and Qwen3 architectures.
• Standard RoPE: Full rotary embeddings on all 128 head dimensions, standard LLaMA-style half-split convention.
• Lower throughput expected: 64 experts with top-8 routing results in significant compute per token despite only 1B active parameters.

Related Issues

N/A

vLLM Integration

• This model/feature is intended for use with vLLM
• Documentation includes vLLM registration instructions

---

By submitting this PR, I confirm that:

• I have read and followed the contributing guidelines
• This is a community contribution and may have limited testing compared to officially-supported models
• The code follows best practices and is well-documented
• All required components listed above are included