Document the current training and vLLM guidance status

docs/issues/issue-2-training-and-vllm.md

@@ -0,0 +1,21 @@
+# Issue #2 — Training script and serving guidance status
+
+## Summary
+The current public repository is focused on inference and model conversion. It does not yet include the minimal training reference requested in issue #2.
+
+## Current repository state
+- The repository includes inference examples and conversion scripts.
+- Public training / fine-tuning scripts are not currently present.
+- In the issue thread, one commenter suggests that `vllm-omni` is the most relevant avenue for multimodal serving.
+- In the same thread, a collaborator states that the team plans to release a minimal version of the training code in the coming weeks.
+
+## What this means today
+- A public minimal training reference is not yet available in the repository.
+- vLLM integration guidance is not yet documented as a first-class supported serving path in the repo itself.
+
+## Safe guidance captured from the issue thread
+- For multimodal serving exploration, `vllm-omni` is the most specific direction mentioned in the thread.
+- For training, the public status is: planned minimal training-code release, but not yet shipped here.
+
+## Why this note exists
+Issue #2 asks for status and guidance. This note records the current public answer without overstating repository capabilities.