opensource/cloudpods
3
0
Fork 0
mirror of https://github.com/yunionio/cloudpods.git synced 2026-09-21 06:09:39 +08:00
Code Issues Packages Projects Releases Wiki Activity
20,956 Commits 27 Branches 687 Tags
d45131bbe0298d91310e984ece03ac44a7c52744
Commit Graph
77 Commits
Author SHA1 Message Date
Zexi Li d45131bbe0 feat(llm,host): support Hygon DCU runtime for vLLM containers (#25258) 
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 13:51:44 +08:00
Zexi Li b936bb2f40 feat(host): support Hygon DCU isolated device and HAMI (#25245) 
Add Hygon DCU container device passthrough, HAMI vGPU sharing, vendor
field/filter on isolated devices, remote path helper, and LLM SKU mapping.
 2026-07-29 19:01:33 +08:00
Zexi Li 0fa97910f3 fix(llm): require memory_mb for local_path HAMi devices (#25220) 
local_path SKUs cannot estimate VRAM from InstantModel; validate HAMi devices set devices[].memory_mb on create/update and deployment.
 2026-07-29 10:21:57 +08:00
Zexi Li a1012f738a feat(llm): support hami (#25215) 2026-07-23 17:08:37 +08:00
屈轩 5f0cb67c4d fix(mcp-server): optimized mcp desc (#25194) 2026-07-21 16:19:25 +08:00
屈轩 21822dc778 fix(mcp-server): mcp use climc struct (#25184) 2026-07-20 17:17:24 +08:00
wanyaoqi 574cac011a feat(region,host,scheduler): isolated device refactor and hami support (#25129) 2026-07-20 13:04:51 +08:00
Zexi Li 98d4515548 feat(aiproxy): auto-provision placeholder ai_key for upstream resolve (#25155) 2026-07-15 19:24:43 +08:00
cwz_eikoh e6c6f66760 feat(llm): add llm-router-agent (#25149) 2026-07-15 15:32:45 +08:00
cwz_eikoh f3bc838d41 feat(llm): support llm-bench in llm (#25147) 2026-07-15 15:03:15 +08:00
Zexi Li 0d58ed4ccb feat(aiproxy): add OpenAI Responses API with provider adapters and Codex support (#25121) 
Implement /v1/responses endpoint with multi-provider translation (OpenAI, Azure,
Anthropic, Gemini), streaming compat layers, Codex config/tool integration, and
updated functional test docs.
 2026-07-10 16:40:42 +08:00
Zexi Li f538dcf2e6 feat(aiproxy): add Anthropic Messages API and migrate functional tests to Go (#25097) 
Add /v1/messages handler with Anthropic-to-OpenAI translation, upstream
failover, and probe endpoints. Replace shell-based functional test scripts
with pkg/aiproxy/ft and climc test commands; consolidate documentation.
 2026-07-07 20:13:03 +08:00
Zexi Li c91b5bc70c feat(llm): add ModelScope instant model import and refactor model download (#25091) 
Add ModelScope hub integration, download driver, and instant model import
flow parallel to HuggingFace. Extract shared HuggingFace download helpers
and reuse them across vLLM/SGLang container drivers.
 2026-07-01 20:12:59 +08:00
Zexi Li f1f3c1df2a feat(llm): support local_path scheduling with prefer_hosts and deployment restart (#25085) 2026-07-01 13:11:02 +08:00
Zexi Li 3413845e96 feat(llm): backfill orphan LLM instances into deployments on init (#25079) 
On service startup, find LLM instances without llm_deployment_id and
create a matching SLLMDeployment, then link the instance. Idempotent
across restarts; skipped on slave nodes.
 2026-06-29 11:26:16 +08:00
Zexi Liandcwz d3b4ab273a Automated cherry pick of #25074: feat(llm): tool call defaults & auto gpu ultilization (#25075) 
* feat(llm): upgrade auto-gpu-ultilization

* feat(llm): add vllm tool call defaults

* feat(llm): add sglang tool call defaults

---------

Co-authored-by: cwz <cwz_eikoh@163.com>
 2026-06-26 10:15:55 +08:00
Zexi Li 686b0f3872 feat(llm): async instant model delete and improve import status tracking (#25072) 
fix(aiproxy): set OpenAI handler process timeouts and skip upstream error
response when request context is already cancelled.
 2026-06-25 19:52:28 +08:00
Zexi Li b52aa5de85 feat(llm): validate GPU devices on deployment create and improve error messages (#25062) 2026-06-24 14:26:18 +08:00
cwz_eikoh c3db76457d Automated cherry pick of #25016: Feature/http get+llm probe (#25045) 
* hostman: probe http get in pod netns

* llm: watch pod status for service readiness
 2026-06-23 19:40:46 +08:00
Zexi Li 4237ce52ea feat(llm): sync llm deployment with aiproxy catalog and routing (#25013) 
Add auto_register_aiproxy on llm_deployment creation, background sync task,
and register/unregister CLI actions. Link aiproxy providers, routings, and
models to llm replicas via llm_id and llm_deployment_id fields.
 2026-06-22 19:35:50 +08:00
cwz_eikoh d2bbda642b Automated cherry pick of #25044: Feature/llm route for aiproxy (#25046) 
* feat(aiproxy): support routing model in aiproxy

* feat(llm): add llm-router
 2026-06-22 10:09:58 +08:00
Zexi Li 81ebea9451 feat(llm): add community images catalog API (#25030) 
Expose llmimages.yaml as list/show/refresh endpoints with configurable
source URL, periodic cache refresh, and mcclient module support.
 2026-06-15 17:26:25 +08:00
屈轩 7cff45fb5b feat(aiproxy): add aiproxy (#25003) 2026-06-09 18:54:54 +08:00
Zexi Liandcwz 72d7bd23b4 Automated cherry pick of #24985: Feat/deployment hostpath (#24993) 
* feat(llm): hostpath support in deployment

* fix(llm): status error in deployment & sku deleted by deployment accidentally

---------

Co-authored-by: cwz <cwz_eikoh@163.com>
 2026-06-08 11:23:48 +08:00
Zexi Li e42070a852 fix(llm): 更新 model catalog 默认 URL 与 webtop 镜像同步 (#24973) 
将 ModelCatalogURL 默认路径改为 model-catalog.yaml；
sync_llm_images 改为同步 fedora-xfce webtop 并注释其它候选镜像。
 2026-06-04 18:51:12 +08:00
Zexi Li a1ba61b875 fix(llm): container desktop (#24970) 2026-06-04 11:51:41 +08:00
Zexi Li e3fb9b2295 feat(llm): container desktop (#24961) 2026-06-03 13:24:53 +08:00
cwz_eikoh 28b9e2e155 Automated cherry pick of #24924: Feat/auto gpu ultilization (#24929) 
* fix(llm): support hf import llm_sku

* feat(llm): auto-gpu-memory-utilization
 2026-05-29 17:45:48 +08:00
Zexi Liandcwz d87805622e Automated cherry pick of #24885: Feature/download progress (#24909) 
* feat(llm): support download progress when model import

* fix(llm): unsupported catalog source \"ollama\" when model_set import llm_sku

---------

Co-authored-by: cwz <cwz_eikoh@163.com>
 2026-05-27 13:20:54 +08:00
Zexi Li 4c0dd07a91 feat(llm,scheduler): vram calculation (#24884) 2026-05-25 16:27:20 +08:00
Zexi Liandcwz ecdf0e824f feat(llm): auto append backend_parameters in vllm & sglang (#24883) 
Co-authored-by: cwz <cwz_eikoh@163.com>
 2026-05-22 17:02:18 +08:00
Zexi Liandcwz 7ed23c433d feat(hermes): connect vllm/sglang/ollama (#24863) 
Co-authored-by: cwz <cwz_eikoh@163.com>
 2026-05-21 11:41:22 +08:00
Zexi Liandcwz c019216508 feat(llm): create llm_sku from llm_model_set (#24865) 
Co-authored-by: cwz <cwz_eikoh@163.com>
 2026-05-21 11:40:37 +08:00
Zexi Liandcwz 383764dbb6 feat(llm): Implement resumable downloads for models from Hugging Face (#24859) 
Co-authored-by: cwz <cwz_eikoh@163.com>
 2026-05-21 10:48:39 +08:00
Zexi Li 9775565a53 feat(llm): llm deployemnt (#24847) 2026-05-20 11:06:13 +08:00
Zexi Li d38b835f8e fix(llm): persistent /config dir of hermes agent (#24834) 2026-05-18 14:57:26 +08:00
Zexi Liandcwz 41a5d7aa3f feat(llm): support sglang (#24822) 
Co-authored-by: cwz <cwz_eikoh@163.com>
 2026-05-14 17:07:37 +08:00
Zexi Liandcwz 520bab0dca Automated cherry pick of #24801: Feature(vllm): use entrypint instead of StartLLM (#24803) 
* feat(llm): use entrypoint instead of StartLLM

* feat(llm): adapt front-end in comfyui instant-model

---------

Co-authored-by: cwz <cwz_eikoh@163.com>
 2026-05-12 10:46:03 +08:00
Zexi Li 48fcab103c fix(llm): honor SLLM device override when building container spec (#24798) 2026-05-11 20:06:12 +08:00
Zexi Li f4503e2284 feat(llm): add SLLM mounted_models override and driver-level validation (#24795) 2026-05-11 17:41:05 +08:00
Zexi Liandcwz dc4fd66eba Automated cherry pick of #24760: Feature/comfyui instant model (#24775) 
* feat(llm): support cursor

* feat(llm): support import from huggingface

* fix(llm): direction error

* feat(llm): support instant model for comfyui

* feat(llm): support multi file select when save-instant-model

---------

Co-authored-by: cwz <cwz_eikoh@163.com>
 2026-05-07 14:42:53 +08:00
wanyaoqi e9dde8eae6 fix(mcclient,region): add tasks mod, service app options (#24734) 2026-04-29 23:48:06 +08:00
Zexi Li 0f30ccf520 fix(host,llm): fix pod shm mount path return and llm restart memory config (#24737) 
- Return shmPath instead of empty string when mountpoint is already
  mounted, so callers get the correct path
- Remove incorrect +1 offset in memory comparison and size formatting
  during LLM restart config change
 2026-04-28 19:57:42 +08:00
Zexi Li 9be35e4b45 feat(llm): support OpenClaw manual config mode (#24719) 
When ManualConfig is enabled, inject OPENCLAW_MANUAL_CONFIG=1 env var
and skip provider/channel credential injection, allowing users to
configure providers and channels themselves inside the desktop.
 2026-04-24 11:45:46 +08:00
Zexi Liandcwz 67cceae9e2 Automated cherry pick of #24717: Feature/hf-search (#24718) 
* llm: add instant-model huggingface query commands

* fix(vllm): use ping instead of health

---------

Co-authored-by: cwz <cwz_eikoh@163.com>
 2026-04-24 10:28:34 +08:00
Zexi Li 6372ef9819 refactor(llm): extract shared desktop webtop config (#24699) 
Move common desktop/webtop container envs, base spec, rootfs, and login-info parsing
into llm_container helpers and reuse in openclaw and hermes-agent drivers.
 2026-04-21 15:04:44 +08:00
Zexi Li 23e21c2c16 feat(llm): support hermes-agent (#24696) 2026-04-20 12:47:23 +08:00
Zexi Liandcwz 2ee8ae31ad feat(llm): add hostpaths for llm-sku (#24695) 
Co-authored-by: cwz <cwz_eikoh@163.com>
 2026-04-20 10:51:40 +08:00
Zexi Li 218c973638 fix(llm): harden volume resize and update size log (#24631) 
- Wait for disk/desktop status after resize via backup worker
- Centralize volume SizeMB update with resize action log
- Expose vCPU/memory fields in LLM customize columns
 2026-04-13 14:38:50 +08:00
Zexi Liandcwz dffac28e32 feat(llm): get available models from base_url (for ollama & openai) (#24608) 
Co-authored-by: cwz <cwz_eikoh@163.com>
 2026-04-01 15:44:30 +08:00