Align the chunked-loss config access between SFT and Distillation

[qgallouedec]

Both trainers resolve the same per-model logit post-processing (logit_scale / output_multiplier / final_logit_softcapping) for their chunked loss, but reached it two different ways. DistillationTrainer uses config.get_text_config(); SFT branched on an is_vlm flag.

• _patch_chunked_ce_lm_head now reads model.config.get_text_config(), so the resolution block is word-for-word identical to the one in distillation_trainer.py.
• is_vlm stays — it still gates two transformers < 5.0.0 fallbacks (VLMs set base_model_prefix = "", and take config-level MoE aux-loss parameters) — but its docstring no longer claims a config-reading role it no longer has.
• TestPatchChunkedCELMHead + TestChunkedCrossEntropyLoss: 79 passed.

---

Note

Low Risk
Narrow config-resolution change in chunked CE training path; behavior should match transformers’ canonical get_text_config() and existing distillation logic, with tests reported passing.

Overview
_patch_chunked_ce_lm_head now resolves logit_scale, output_multiplier, final_logit_softcapping, and MoE output_router_logits via model.config.get_text_config(), matching DistillationTrainer instead of branching on is_vlm (text_config vs top-level config).

The is_vlm argument is unchanged for transformers < 5.0.0 (backbone via model.model, config-level MoE aux-loss fields); its docstring no longer describes config selection. Inline comments were updated to document get_text_config() and the logit_scale / output_multiplier fallback behavior.

^{Reviewed by Cursor Bugbot for commit 834ec80. Bugbot is set up for automated code reviews on this repo. Configure here.}

[bot-ci-comment]

The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update.