transformers

7cf4241f - Add expert parallelism (EP) config support for Qwen3 MoE (#45436)

Add expert parallelism (EP) config support for Qwen3 MoE (#45436) * Add base_model_ep_plan to Qwen3OmniMoeTalkerTextConfig Add base_model_ep_plan to Qwen3VLMoeTextConfig * Add expert-only EP plan to Qwen3 MoE configs Defines sharding strategy for MoE experts without affecting attention layers, allowing EP to scale beyond num_kv_heads constraints. * Use expert-only EP plan for Qwen3 MoE configs Remove duplicate base_model_ep_plan with attention entries from qwen3_moe and update qwen3_vl_moe to use the expert-only EP plan. Attention is left unsharded — FSDP2 handles attention weight distribution. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com> * Remove comments from generated qwen3_vl_moe config * Apply repo consistency fixes * Add base_model_ep_plan to qwen3_vl_moe and qwen3_omni_moe modular files and regenerate configs * Fix base_model_ep_plan position in generated configs to match modular converter output --------- Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com> Co-authored-by: github-actions[bot] <github-actions[bot]@users.noreply.github.com>