aitfSR4/ub-sr04-qwen3.5-4b-cpt2-sft-game

Upload 2 files

fix: patch hidden_size to root level for vLLM compatibility

Fix config.json: correct hidden_size=2560 from Qwen3.5-4B-Base

Fix model_type: qwen3_5_text -> qwen3_5 (vLLM compatible)

Re-export: clean bfloat16, Base+CPT+SFT merged properly (no quantization)

Fix config: remove quantization_config (model is bfloat16, vLLM compatible)

Replace with bfloat16 merged model (vLLM compatible, no bitsandbytes)

Add merged CPT+SFT model weights (Qwen3.5-4B-Base + CPT Tim2 + SFT Tim4)

Add merged CPT+SFT model (Qwen3.5-4B-Base + CPT Tim2 + SFT Tim4)

initial commit