Static INT8 audio encoder (Entropy calibration, MatMul-only, QOperator format — browser-compatible)

Add static INT8 audio encoder (QOperator format, browser-compatible)

Add INT8 audio encoder

Add tokenizer.json (transformers.js fast tokenizer)

Update README: GPTQ-calibrated INT4 (92.7%)

GPTQ INT4 decoder_step_int4.onnx.data

GPTQ INT4 decoder_step_int4.onnx

GPTQ INT4 decoder_prefill_int4.onnx.data

GPTQ INT4 decoder_prefill_int4.onnx

Update quality table with mixed8 (INT8 enc + INT4 dec = 91.9%) as recommended config

Update manifest: add forced-language prompts (default: English)

Add pre-computed mel features for the 8 VITW examples + prompt manifest

Add embedding table (onnx/embed_int8_scales.bin)

Add embedding table (onnx/embed_int8.bin)

Add README with full breakdown, tags, base_model_relation: quantized

Initial upload: INT4 ONNX (encoder + prefill + step) + tokenizer + examples + inference.py

initial commit