wehub-resource-sync
81 lines
1.9 KiB
ReStructuredText
81 lines
1.9 KiB
ReStructuredText
.. _recipe_devstral:
|
|
|
|
Mistral / Devstral
|
|
==================
|
|
|
|
Validated models
|
|
----------------
|
|
|
|
- `mistralai/Devstral-2-123B-Instruct-2512 <https://huggingface.co/mistralai/Devstral-2-123B-Instruct-2512>`_
|
|
|
|
.. tab-set::
|
|
:sync-group: engine
|
|
|
|
.. tab-item:: vLLM
|
|
|
|
**Engine documentation:**
|
|
`Mistral / Devstral in vLLM supported models
|
|
<https://docs.vllm.ai/en/latest/models/supported_models.html#text-generation>`_
|
|
(architecture ``MistralForCausalLM``).
|
|
|
|
**Status:** Validated with LMCache.
|
|
|
|
Start the LMCache MP server:
|
|
|
|
.. code-block:: bash
|
|
|
|
lmcache server --l1-size-gb 100 --eviction-policy LRU
|
|
|
|
|
|
|
|
|
Start vLLM with the LMCache MP connector:
|
|
|
|
.. code-block:: bash
|
|
|
|
vllm serve mistralai/Devstral-2-123B-Instruct-2512 \
|
|
--tensor-parallel-size 2 \
|
|
--enable-auto-tool-choice \
|
|
--tool-call-parser mistral \
|
|
--kv-transfer-config \
|
|
'{"kv_connector":"LMCacheMPConnector", "kv_role":"kv_both"}'
|
|
|
|
|
|
|
|
|
Adjust ``--tensor-parallel-size`` to match your hardware. For the
|
|
generic LMCache + vLLM wiring (ports, remote hosts),
|
|
see :doc:`../getting_started/quickstart`.
|
|
|
|
If there are any issues with vLLM setup, please refer to the
|
|
`vLLM Recipes <https://docs.vllm.ai/projects/recipes/en/latest/index.html>`_
|
|
for more details.
|
|
|
|
.. tab-item:: SGLang
|
|
|
|
**Status:** Not validated with LMCache.
|
|
|
|
.. tab-item:: TRT-LLM
|
|
|
|
**Status:** Supported. See :doc:`../getting_started/quickstart` for TRT-LLM + LMCache setup.
|
|
|
|
CacheBlend support
|
|
------------------
|
|
|
|
Compression support
|
|
-------------------
|
|
|
|
.. list-table::
|
|
:header-rows: 1
|
|
:widths: 25 20 55
|
|
|
|
* - Method
|
|
- Status
|
|
- Notes
|
|
* - :doc:`CacheGen <../kv_cache_optimizations/compression/cachegen>`
|
|
- Not validated
|
|
-
|
|
|
|
Caveats
|
|
-------
|
|
|
|
None known.
|