0.1.0

✨ Unified encode API across transformer models

The transformer model bundles — Bert, ModernBert, NomicBert, Qwen3, Roberta, XLMRoberta, and now Clip — share a single EncodeOptions struct on encode / batchEncode:

var options = EncodeOptions.bert
options.postProcess = .meanPool(normalize: true)
options.computePolicy = .cpuOnly
let encoded = try modelBundle.encode("The cat is black", options: options)

EncodeOptions consolidates the previously per-model parameters:

• maxLength — max tokens per input sequence
• padTokenId — padding id for batchEncode (nil → model's own id)
• postProcess — pooling strategy
• computePolicy — MLComputePolicy for the underlying MLTensor work

🤖 Pooling & max-length derived from sentence-transformers config

When you don't pass options, each bundle uses its defaultEncodeOptions, now derived at load time from a model's sentence-transformers layout when present:

• modules.json
• 1_Pooling/config.json
• sentence_*_config.json (transformer config)
• tokenizer model_max_length

This means sentence-transformers models are pooled correctly out of the box — e.g.:

• nomic-ai/nomic-embed-text-v1.5 → .meanPool(normalize: false)
• nomic-ai/modernbert-embed-base → .meanPool(normalize: true)

For plain checkpoints with no such config, bundles fall back to a per-model static default (.bert, .modernBert, .nomicBert, .qwen3, .roberta, .xlmRoberta, .clip).

🔧 Other changes

• Clip adopted EncodeOptions for a consistent API; its pooling stays fixed (EOS token → projection → L2 normalize), so postProcess is ignored.
• CLI commands updated to the unified API.
• README updated to document EncodeOptions, pooling, and the sentence-transformers auto-derivation.

⚠️ Breaking changes

• encode / batchEncode on the transformer (and Clip) bundles now take a single options: EncodeOptions? argument instead of individual parameters (maxLength, padTokenId, postProcess, computePolicy). Migrate by constructing an EncodeOptions (or starting from a static default / bundle.defaultEncodeOptions) and setting the fields you need.
• Embedding-table bundles (Model2Vec, StaticEmbeddings) and Word2Vec are unchanged.

Full Changelog: 0.0.30...0.1.0