Add ArlowGPT

[yuchenxie4645]

Add Arlow Model & Tokenizer Support

Adds support for the Arlow model and its corresponding ArlowTokenizer.

Features:

• Flash Attention 2 for fast memory-efficient training
• Rotary Position Embeddings (RoPE) with rope_theta=100000.0
• Grouped Query Attention (GQA)
• Cross-Attention for future multimodal extensions
• RMSNorm, SiLU activations, tied embeddings
• Supports full causal language modeling (ArlowForCausalLM)
• ArlowTokenizerFast (fast tokenizer, vocab size 131072)
• ArlowTokenizer (Non fast tokenizer)

Includes:

• ArlowConfig, ArlowModel, ArlowForCausalLM, ArlowPreTrainedModel
• ArlowTokenizerFast as fast tokenizer (tokenization_arlow_fast.py)
• ArlowTokenizer as fast tokenizer (tokenization_arlow.py)
• Auto mapping + lazy loading registration

@ArthurZucker

PR Link: Here

Open source status

• The model implementation is available
• The model weights are available

Provide useful links for the implementation

#36899