* feat: update llama.cpp to 6eab47181

* feat: add Pyodide wheel support

* docs: fix Pyodide changelog entry

* feat: enable mtmd for emscripten

…ready cached (#2306)

Co-authored-by: Ankur Kaul <akaul36@gatech.edu>

* feat: update llama.cpp to fdbd6abee

* feat: update llama.cpp to b3fed31b9

* feat: update llama.cpp to 78d2f5246

* fix: align llama.cpp binding comments

* fix: align llama.cpp binding docstrings (#2328)

* fix: align llama.cpp binding docstrings

* fix: use None in Python binding docstrings

* fix: format binding docstring

* feat: update llama.cpp to 13f2b28b0

* feat: update llama.cpp to e3546c794

* feat: update llama.cpp to d2a818231

* feat: update llama.cpp to 11924d4c1

* fix: close contexts before models

* feat: update llama.cpp to fc3f10b38

* feat: update llama.cpp to 42e98813e

* feat: update llama.cpp to 936918514

* fix(server): show falsey defaults in CLI help

* test(server): remove CLI regression test

---------

Co-authored-by: haijie11 <haijie11@noreply.gitcode.com>
Co-authored-by: abetlen <abetlen@gmail.com>

) (#2343)

* fix(llama): release a vision chat handler's mtmd context on close (#2342)

`Llama.close()` tore down the model and its context but left the chat
handler's `_exit_stack` untouched, so the mtmd/clip context it built from
that model was never freed.

Handlers routinely outlive the `Llama` that initialized them -- callers
construct one handler and reuse it across loads -- and
`_init_mtmd_context()` returns early while `mtmd_ctx` is set. After the
first `close()` the handler therefore kept a context bound to an
already-freed model and handed it back on the next load, where it
surfaces as a null `mtmd_ctx` on the C++ side.

Register the handler's exit stack on the Llama's `_stack`. It unwinds
LIFO, so the mtmd context is released ahead of the model teardown
registered earlier in `__init__`, and `mtmd_free` resets `mtmd_ctx` to
`None` so the next load re-initializes cleanly. Duck-typed on
`_exit_stack`, which both `Llava15ChatHandler` and `MTMDChatHandler`
(and every subclass) expose.

* fix: bind multimodal context cleanup to model

---------

Co-authored-by: Anai Guo <antai12232931@anaiguo.com>
Co-authored-by: abetlen <abetlen@gmail.com>