Restore original fused float16 model.safetensors for MLX (4.2GB)

Remove GPTQ model.safetensors.index.json

Remove GPTQ model-00002-of-00002.safetensors

Remove GPTQ model-00001-of-00002.safetensors

v0.5.9: semantic intent routing

v0.5.8: 3-tier inference (MLX/vLLM/llama.cpp)

v0.5.8: GPTQ W4A16 quantized model for vLLM CPU (~4GB)

Delete model.safetensors.index.json with huggingface_hub

Delete model-00004-of-00004.safetensors with huggingface_hub

Delete model-00003-of-00004.safetensors with huggingface_hub

Delete model-00002-of-00004.safetensors with huggingface_hub

Delete model-00001-of-00004.safetensors with huggingface_hub

Delete model.safetensors with huggingface_hub

v0.5.8: Replace MLX-quantized with proper dequantized safetensors for llm-compressor

Fix config.json: remove invalid GGML quantization fields

v0.5.7

Fix tokenizer_class: TokenizersBackend → PreTrainedTokenizerFast

v0.5.6

v0.5.5

v0.5.4: vLLM CPU pre-built wheel, bfloat16, TCMalloc

v0.5.3: fix CPU device detection, consolidate changelog

Update README to v0.5.2

Update README to v0.5.1

Initial release v0.5.0 (Clean history)