xLLM Part I model artifacts

The model weights are distributed under the Apache License, Version 2.0.
A copy is provided in LICENSE. The xLLM code has its own license in the
repository root; third-party code retains its original notices.

Tokenizer attribution

The tokenizer is derived from IFM/K2-Horizon-3.7B and retains its upstream
Apache-2.0 license.
Source revision: 86a26d7bf092484faa047121d5816f037d4a8ffa
https://huggingface.co/IFM/K2-Horizon-3.7B/tree/86a26d7bf092484faa047121d5816f037d4a8ffa

Part I modifications:
- tokenizer.json retains the final ByteLevel post-processor from the source
  Sequence, removing automatic begin-of-text insertion.
- special_tokens_map.json stores each token object's content as a string.
- tokenizer_config.json is copied unchanged.
The vocabulary, token IDs, merges and added-token attributes are unchanged.

Artifact preparation:
- Source model tensors are converted to BF16 and stored in Safetensors shards.
- config.json holds the checkpoint's values of the model fields its recipe
  sets and the tokenizer settings; tokenizer.path points to this artifact's
  tokenizer directory.