Update README.md

README.md

@@ -20,10 +20,6 @@ Qwen3.5 models are natively multimodal (VLM). Their HuggingFace checkpoints use
 - Weight keys at `model.layers.*` (standard causal LM format, no `language_model.` prefix)
 - Vision encoder and MTP weights removed
 
-## Why this exists
-
-When training frameworks like TRL save Qwen3.5 text-only checkpoints during GRPO/RL training, they produce this format. vLLM needs a public checkpoint in this format for CI testing of the `Qwen3_5ForCausalLM` code path. See [vllm-project/vllm#36275](https://github.com/vllm-project/vllm/issues/36275).
-
 ## Model structure
 
 - **Architecture**: Hybrid GatedDeltaNet (24 layers) + Full Attention (8 layers)