convert : fetch tokenizer vocab and remote code with --remote

[GrEarl]

--remote currently downloads ["LICENSE", "*.json", "*.md", "*.txt", "tokenizer.model"]. Two kinds of tokenizer files are not covered, so set_vocab() fails for models that need them even though the same model converts fine from a full local snapshot:

• tiktoken.model — where Kimi K2 / K2.5 / K3 keep their vocabulary. Not matched by tokenizer.model.
• *.py — several converters call AutoTokenizer.from_pretrained(self.dir_model, trust_remote_code=True), which resolves the tokenizer class from the model directory. For Kimi that is tokenization_kimi.py, which in turn relative-imports encoding_k3.py.

Concretely, on moonshotai/Kimi-K3 via kimi_linear.py:

OSError: ... does not appear to have a file named tokenization_kimi.py

deepseek.py reaches the same call for Kimi-K2/K2.5, so this is not specific to one architecture.

Measured on moonshotai/Kimi-K3 with snapshot_download:

	files	size	tiktoken.model	tokenization_kimi.py	encoding_k3.py
before	15	57.05 MiB	missing	missing	missing
after	33	59.89 MiB	ok	ok	ok

+2.84 MiB. Most of the 57 MiB in both cases is model.safetensors.index.json, already pulled by *.json.

*.model subsumes the existing tokenizer.model entry, so that one is dropped rather than kept alongside.

On *.py: this only makes --remote match what already happens for a local directory. The converters that need these files already run them via trust_remote_code=True; today the download simply omits them and the conversion aborts. If pulling model code unconditionally is not wanted, the alternative is to pass the already-available self.remote_hf_model_id to AutoTokenizer.from_pretrained so transformers resolves the code from the Hub, but that touches every converter that makes the call rather than one line here.

Not verified: I have not run a full --remote conversion to completion on a model that needs this — the download side is what I measured. Found while testing #26185 (model: add Kimi-K3 text model), where --remote stops in set_vocab for this reason; details in #26185 (comment)

AI usage disclosure: yes. The before/after file listing above is output from a command I ran; the change is the one line shown in the diff.

[ggml-gh-bot]

Hi @GrEarl, thanks for your contribution!

Per our contribution guidelines, the automated PR checker found the following issue(s) that need your attention:

• PR Template not respected: Please respect the template when creating a new pull request. Make sure to fill out all required sections.

• Multiple open PRs from a new contributor: We limit new contributors (those without a previously merged PR) to 1 open PR at a time. You currently have 2 open PRs.

• AI-generated content: While code is allowed to be generated by AI, please write the PR description and commit messages on your own without the help of AI.

---

Please note that maintainers reserve the right to make final decisions on PRs. If you believe there is a mistake, please comment below.