v0.6.0

What's Changed

• Adopt llamawasm2go v0.5.0 and add ModernBERT encoder tests by @goccy in #22

Full Changelog: v0.5.1...v0.6.0

v0.5.1

What's Changed

• Safely close contexts and join worker threads by @goccy in #21

Full Changelog: v0.5.0...v0.5.1

v0.5.0

What's Changed

• Adopt llamawasm2go v0.3.5: native kernels for every quant path, tensor-to-kernel coverage test by @goccy in #18
• test: q5_K tile identity on every engine path; sliced-vs-whole scoring bound by @goccy in #19
• Adopt llamawasm2go v0.4.0: Slots (continuous batching), KVUnified, lossless text by @goccy in #20

Full Changelog: v0.4.3...v0.5.0

v0.4.3

What's Changed

• Adopt llamawasm2go v0.3.4: finite scores on AVX-512 VNNI hosts by @goccy in #17

Full Changelog: v0.4.2...v0.4.3

v0.4.2

What's Changed

• Adopt llamawasm2go v0.3.3: Score past n_batch, wikitext-2 perplexity parity, CPU-keyed perf baseline by @goccy in #16

Full Changelog: v0.4.1...v0.4.2

v0.4.1

What's Changed

• Adopt llamawasm2go v0.3.2 and the llama-wasm v0.3.2 bridge by @goccy in #14
• make: refuse a Rosetta amd64 toolchain; add test-arm64-native by @goccy in #13
• ci: gate throughput vs native llama.cpp on arm64 and amd64 (10% on the go/native ratio) by @goccy in #15

Full Changelog: v0.4.0...v0.4.1

v0.4.0

What's Changed

• Add Context.ScoreChoices: batched candidate scoring over the cached stem by @goccy in #11

Full Changelog: v0.3.0...v0.4.0

v0.3.0

What's Changed

• Add instance snapshots: NewSnapshot / Snapshot.Fork by @goccy in #10

Full Changelog: v0.2.3...v0.3.0

v0.2.3

What's Changed

• Build the shared images in place and adopt the llamawasm2go v0.2.4 engine by @goccy in #9

Full Changelog: v0.2.2...v0.2.3

v0.2.2

What's Changed

• Adopt the llamawasm2go v0.2.3 engine bundle: transpiler-guarded spin loops by @goccy in #8

Full Changelog: v0.2.1...v0.2.2