Add ROCM_ACCELERATION.md — deep dive into ROCm model acceleration libraries and how they relate to StreamingVLM

Add EXPLAINED.md — deep dive into StreamingVLM architecture and patching methodology

Update ML Intern artifact metadata

Update README: add Qwen3-VL-2B support, CUDA flash_attention_2, dual platform guide

Add Qwen3-VL-2B support + CUDA flash_attention_2 config

Add MIGRATION_GUIDE.md

Add DeepSpeed ZeRO-3 config

Add stage 2 script and zero3 config

Add training scripts and DeepSpeed config

Add inference.py — main streaming loop

Add get_qwen_range.py token utilities

Add pos_emb.py — contiguous RoPE for infinite streams

Add model_forward.py — top-level streaming forward

Add vision_forward.py — chunked SDPA for vision encoder

Add language_forward.py — SDPA attention for text decoder

Add qwen3 patch_model.py

Add streaming_cache.py

Add streaming_args.py

Add utils init

Add generate init

Add qwen3 and generate init files

Add inference subpackage init files

Add streaming_vlm package init files

Add test_imports.py

Add train.py and test_imports.py

Add setup.py

Add requirements, setup, and training files

Update ML Intern artifact metadata

Initial commit: StreamingVLM port for Qwen3-VL 4B on ROCm

Update ML Intern artifact metadata

initial commit