Annotate dsv3 with layer_id

[SherlockNoMad]

Stack from ghstack (oldest at bottom):

• -> Annotate dsv3 with layer_id #1908
• Annotate EP with dispatch/compute/combine #1907
• Setup Compiler Toolkit experiment folder for dsv3 #1906

Sample output: P1996285071

[wwwjn]

Can you help me understand what does these notation for? And it's for joint_graph_runner, right? If we need more notations on the model in the future, it's better to keep a separate copy of Deespeek-v3 model (like simple_fsdp: https://github.com/pytorch/torchtitan/blob/main/torchtitan/experiments/simple_fsdp/deepseek_v3/model.py), so it won't confuse the majority of titan users.

[SherlockNoMad]

Can you help me understand what does these notation for? And it's for joint_graph_runner, right? If we need more notations on the model in the future, it's better to keep a separate copy of Deespeek-v3 model (like simple_fsdp: https://github.com/pytorch/torchtitan/blob/main/torchtitan/experiments/simple_fsdp/deepseek_v3/model.py), so it won't confuse the majority of titan users.

yes, it's for joint graph runner.
We will make the clone of the model.