docs: 벤치마크 섹션 한국어로 변환

data: add raw benchmark results JSON (KMMLU + HAE-RAE, 3-model comparison)

docs: add Korean LLM benchmark results (KMMLU + HAE-RAE, 3-model comparison)

fix: replace Qwen default system prompt with VELA identity in chat templates

docs: add llama-cpp-python, vllm, ollama tags for library discovery

docs: add MLX 4-bit format, update quant links in model card

feat: add MLX 4-bit quantized model (Apple Silicon optimized)

docs: update README to v1.3 (DPO v6)

feat: update to DPO v6 merged model (BF16 safetensors)

feat: add DPO v6 GGUF Q4_K_M model

chore: remove v4 GGUF (replacing with v6)

Update README.md

feat: vela-v5-merged 업로드 (SFT merged, rep_penalty=1.05)

fix: generation_config rep_penalty=1.05, top_k=20, top_p=0.8 (vela-v5-merged)

docs: add recommended inference settings and backend configuration guide

fix: match llama-cpp-python defaults (top_k=40, top_p=0.95, rep_penalty=1.0)

fix: rollback generation_config to safe Qwen2.5 defaults (rep_penalty 1.3→1.1)

fix: update generation params — repetition_penalty 1.05→1.3, top_k 20→50, top_p 0.8→0.92

docs: add GitHub framework link and badges

Update README.md

docs: add real output example (RT + Quick Assessment + Analysis Report)

docs: update model card with v3 training data (58K SFT, 26K DPO), MLX benchmark, Markdown RT format

feat: update to DPO v4 merged model (SFT + DPO v4 language leak fix)

Delete vela-q8_0.gguf with huggingface_hub

Delete model.safetensors with huggingface_hub

feat: replace SFT-only GGUF with DPO v4 merged Q4_K_M

Delete vela-q4_k_m.gguf with huggingface_hub

fix: change tokenizer_class to Qwen2TokenizerFast for vLLM compatibility

docs: update training data distribution with accurate numbers (SFT 36,713 + DPO 24,779)

feat: add Q8_0 GGUF quantized model (7.6GB)

feat: add Q4_K_M GGUF quantized model (4.4GB)

docs: update model card with GGUF formats, benchmarks, usage examples

Fix tokenizer_config.json - remove extra_special_tokens list causing vLLM error

Fix config.json for vLLM compatibility (remove layer_types, fix rope_parameters)

Fix year to 2026

Update model card for DPO v4

Update to DPO v4 merged model

Upload folder using huggingface_hub

initial commit