Add mm data example

[yubofredwang]

This pull request updates the dataset configuration and documentation to support OpenAI-style conversations with multimodal and tool-call capabilities. It also adds new sample data covering a range of scenarios, and clarifies the expected dataset format for Kimi-K2.5 training setups.

Dataset format and documentation updates:

• Added detailed dataset format explanations and multimodal conversation examples to examples/kimi-k25-2node-h200/README.md and examples/kimi-k25-3node-h100/README.md, including guidance on using <think>...</think> blocks, image content, and tool calls. [1] [2]
• Clarified that each dataset row should contain a conversations field, and explained how multimodal content is handled for SGLang engine training. [1] [2]

Configuration changes for dataset compatibility:

• Changed the prompt_key from messages to conversations in both configs/sglang_kimi_k25_2node.yaml and configs/sglang_kimi_k25_3node.yaml to match the new dataset format. [1] [2]

Sample data additions:

• Added new samples to examples/data/sample_conversations.jsonl covering text-only, multimodal (single/multi-image), system messages, multi-turn, and tool calls, providing comprehensive examples for model training.

[Copilot]

Pull request overview

This PR updates the Kimi-K2.5 SGLang training examples to use an OpenAI-style conversations column (including multimodal and tool-call examples), and aligns the provided example configs and sample dataset accordingly.

Changes:

• Documented the expected conversations JSONL dataset format (including multimodal + tool-call examples) in the 2-node and 3-node Kimi-K2.5 READMEs.
• Updated Kimi-K2.5 SGLang example configs to use prompt_key: conversations (instead of messages).
• Added new kimi_* example rows to examples/data/sample_conversations.jsonl covering text, images, multi-image, multi-turn, and tool calls.

Reviewed changes

Copilot reviewed 5 out of 5 changed files in this pull request and generated 4 comments.

Show a summary per file

File	Description
examples/kimi-k25-3node-h100/README.md	Adds dataset format guidance and a multimodal OpenAI-style example for the 3-node setup.
examples/kimi-k25-2node-h200/README.md	Adds dataset format guidance and a multimodal OpenAI-style example for the 2-node setup.
examples/data/sample_conversations.jsonl	Appends new kimi_* sample conversations including multimodal and tool-call scenarios.
configs/sglang_kimi_k25_3node.yaml	Switches dataset prompt_key to conversations to match the documented dataset schema.
configs/sglang_kimi_k25_2node.yaml	Switches dataset prompt_key to conversations to match the documented dataset schema.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.