2de295222d
feat(llm): merge SKU envs into container and validate prefer_hosts updates ( #25389 )
...
Allow local_path SKUs to update prefer_hosts on edit, and overlay SKU envs onto the primary container with same-key override.
2026-08-21 14:27:58 +08:00
a13a322bf2
feat(llm): sync isolated devices by sharing_mode on restart ( #25307 )
...
Respect sharing_mode when attaching/restoring GPU devices in compute,
and sync LLM guest devices with SKU config instead of only detaching
when empty.
2026-08-12 20:17:22 +08:00
e7edd821c1
fix(llm): detach stale GPUs when effective devices are empty on restart ( #25280 )
...
Avoid keeping bound isolated devices after SKU/LLM devices are cleared;
desktop container setup now follows effective devices only.
2026-08-06 10:02:22 +08:00
eab7db5518
feat(llm,host): support Hygon DCU runtime for vLLM containers ( #25260 )
...
Wire vendor on LLM SKU devices, resolve supplemental groups for DCU
access, source DTK env in vLLM entrypoint, and add hysmi telegraf metrics.
2026-08-02 13:52:16 +08:00
3431c10196
feat(llm): support local_path scheduling with prefer_hosts and deployment restart ( #25090 )
2026-07-01 13:11:19 +08:00
6c4d68f3b8
fix(llm): honor SLLM device override when building container spec ( #24799 )
2026-05-11 20:06:35 +08:00
b2ec881a68
feat(llm): add SLLM mounted_models override and driver-level validation ( #24797 )
2026-05-11 17:41:28 +08:00
ff89a3839b
feat(llm): add hostpaths for llm-sku ( #24693 )
2026-04-20 10:51:16 +08:00
04d29503ee
feat(llm): 调整 openclaw 容器与 LLM 基础模型/sku 逻辑 ( #24455 )
2026-03-16 16:20:01 +08:00
8b10d33682
Fix(llm): some mistakes, use CmpId & InstantModelId insteadof SvrId & ModelId ( #24194 )
...
* fix(llm): some mistakes
* fix(llm): use CmpId instead of SvrId
* fix(llm): instant_model use id instead of modelId
2026-02-03 15:15:15 +08:00
e254c2ccf5
Fix(llm): fix bugs & apikey encrypt & support history for mcp-agent-chat & delete modelname in llmSku ( #24149 )
...
* fix(llm): mcp-agent apiKey encrypt
* fix(llm): error-handle when instant-model import
* feature(llm): add get-available-network for llm
* feature(llm): support history for mcp-agent-chat
* fix(llm): llm create with net instead of networktype & networkid
* fix(llm): delete modelName in llmSku & fix some bugs
2026-01-29 10:50:00 +08:00
1fc8219ee4
feature(llm): adapt frontend ( #23906 )
...
* feature(llm): add llm-list details, add autostart for llm-save-instant-model
* fix(llm): adjust some interfaces
* fix: name-dup problem when create llm
* fix: install instant-model by id rather than modelID
* fix(llm): add llm_id for mcp-agent
* feature(llm): move network from sku to instance
* feature(llm): add LLMType for llm-image
* feature(llm): add gpuMemoryRequired & ollama-registry yaml
* feature(llm): add url-get interface
* feature(llm): support mcp in mcp-agent-chat
* fix(llm): abstract ollama registry
2026-01-19 18:06:05 +08:00
4c701fa082
Feature/pre quick model ( #23898 )
...
* feat: add preinstall for quick-models
* fix: drop error when duplicate blobs
2025-12-05 10:33:45 +08:00
142e540f7f
[Feature] llm model 使用 post_overlay 功能实现秒装挂载模型 ( #23769 )
...
* feat(llm): add probed-package
* feat(llm): add llm-save-instant-app
* feat(llm): add some llm-instant-app options
* fix(llm): change app to model
* fix(llm): rename origin model to sku
* feat(llm): add llm-quick-models
* fix(llm): saved paths for instant model
* fix(llm): fix quick-install-model
---------
Co-authored-by: Zexi Li <zexi.li@icloud.com >
2025-12-03 16:27:10 +08:00
c414ef4eda
fix(llm): don't rely on polling ( #23708 )
2025-11-07 10:19:30 +08:00
897a6fc61f
[OSPP]feature: Auto deploy llm and dify ( #23282 )
...
* feat(llm): add llm-create
* fix(llm): fix llm-create, add llm-list & llm-show
* fix(llm): format file names
* fear(llm): add exec stream
* fear(llm): init llm model cache
* fix(llm): move llm model cache to LocalImageCache
* feat(llm): auto clean model cache
* fix(llm): auto clean tmp model cache
* feat(dify): init dify-create
* fix(dify): change dify default registry
* feat: support create model with gguf file
* feat(llm): support more modelfile option for gguf file
* feat(dify): init user customize dify parameters
* feat: update llm as a service in climc
* fix: delete llm from region service
* feat: init llm service (with error)
* fix: resolve errors (task cant callback yet)
* feat: add PerformRequestHostActionByOtherService for container
* fix: make llm service usable
* feat: delete llm and dify after guest deleted
* feat(llm): add llm-image
* feat(llm): add llm-model
* feat: mv old llm to ollama
* feat(llm): init llm-batch-create(can not run)
* fix(llm): make llm-create usable
* fix(llm): add pull-model step
* feat(llm): add list and delete
* fix(llm): ollama pull official model don't rely on host & region's code any more
* feat(llm): add dify-model
* fix: remove llm's code in host & compute
* fix: remove remain code in compute
* feat(llm): Abstract out and reuse the llm_model logic
* feat(llm): abstract llm_base from llm
* fix(llm): make dify usable
* feat(llm): add sync_dify_images.sh
* fix(llm): format import
* feat(llm): support start & stop for llm and dify
* feat(llm): add model-update
* fix(llm): gendocgo
2025-10-24 13:29:31 +08:00
Zexi Li