Sirius Claw
240620c3a4
feat: add KEDA scaling for A100 vLLM
...
- Scale to 0 when idle (saves money)
- 15 min cooldown before scale-down
- Triggers on 3+ pending HTTP requests
- Targets openclaw-brain-vllm deployment
2026-04-22 17:07:05 +00:00
sirius0xdev
7041708969
Merge pull request #52 from sirius0xdev/optimize-prompt-context-2026-04-22
...
Optimize prompt context 2026 04 22
2026-04-22 09:58:47 -04:00
Sirius Claw
41f0b43d3b
optimize: trim SOUL.md by ~65%, reduce context injection to curated files only, increase requestTimeout to 180s, enable promptCaching per optimization guide
...
- SOUL.md now <600 bytes (core invariants only)
- Added context.injectAllMarkdown=false + explicit curated list
- requestTimeoutSeconds=180 for high-param models
- promptCaching=true to avoid reprocessing system prompt every turn
- Fixed missing comma in agents.defaults.model (was invalid JSON)
Per https://github.com/OnlyTerp/openclaw-optimization-guide
2026-04-22 12:45:10 +00:00
Sirius Claw
3133a3b32f
fix(vllm): update to recommended qwen3_coder tool parser, auto max-model-len, and latest image
...
Addresses tokenizer backend / parser mismatch on heretic-style Qwen3.5 models per model card recommendations.
2026-04-22 06:32:35 +00:00
sirius0xdev
f435fdaf18
new image
2026-04-22 06:24:17 +00:00
sirius0xdev
c7f0fc7262
fix error
2026-04-22 06:12:32 +00:00
sirius0xdev
c420d2478c
tokenizer mode
2026-04-22 06:09:45 +00:00
sirius0xdev
93caf2d086
image error
2026-04-22 06:01:11 +00:00
sirius0xdev
7864a7f84e
vllm version error
2026-04-22 05:57:45 +00:00
sirius0xdev
8b9e6da69a
config map fix
2026-04-22 05:50:59 +00:00
sirius0xdev
34fe64c74b
model switch again
2026-04-22 05:41:00 +00:00
sirius0xdev
d769356c10
fix config map
2026-04-22 05:28:55 +00:00
Sirius Claw
e7debc708c
Minimal ConfigMap update: Add Qwen3.5-35B-A3B uncensored model as option
...
- Added new model entry to local-vllm.models list (not primary)
- Includes 49k contextWindow and reasoning: true
- No deployment/kustomization changes
2026-04-22 05:03:19 +00:00
sirius0xdev
01a82a74f3
fix typo
2026-04-22 03:40:16 +00:00
sirius0xdev
795e48b9d6
pvc name change
2026-04-22 03:36:55 +00:00
sirius0xdev
1d13cce3cf
switch model
2026-04-22 03:15:29 +00:00
sirius0xdev
1466d38721
fix tokenizer mode
2026-04-22 03:12:58 +00:00
sirius0xdev
d6c50d72ee
fix tokenizer mode
2026-04-22 03:06:16 +00:00
sirius0xdev
6c4bb798ff
switch image
2026-04-22 03:00:46 +00:00
sirius0xdev
c236f3c89a
try and get vllm to work new tags
2026-04-22 02:47:53 +00:00
sirius0xdev
a307e3ae37
fix cuda issue with new image
2026-04-22 02:27:38 +00:00
sirius0xdev
ed096b7cd8
fix vllm image error
2026-04-22 02:07:46 +00:00
sirius0xdev
943efa6b99
turn off keda
2026-04-22 02:04:10 +00:00
sirius0xdev
73c89e361d
reacticate a100
2026-04-22 01:53:36 +00:00
sirius0xdev
e26b7663c6
reactivate a100
2026-04-22 01:53:02 +00:00
sirius0xdev
98d6df80fa
clean up
2026-04-21 23:02:09 +00:00
sirius0xdev
03013de0da
clean up
2026-04-21 23:01:06 +00:00
sirius0xdev
9ce1850986
add xai secret to deployment
2026-04-21 22:56:15 +00:00
sirius0xdev
f888610241
add key
2026-04-21 22:47:40 +00:00
sirius0xdev
c514553c00
add xai_apikey
2026-04-21 22:43:21 +00:00
sirius0xdev
e5f5e7318c
Update kustomization.yaml
2026-04-21 09:55:08 -04:00
sirius0xdev
10d2ff67c1
tokens
2026-04-21 03:55:18 +00:00
sirius0xdev
1d7bb9cba3
fix token length problem
2026-04-21 03:39:54 +00:00
sirius0xdev
fd3e3d8f45
fix openclaw config
2026-04-21 03:28:27 +00:00
sirius0xdev
bd09ea4958
fix oom
2026-04-21 03:03:59 +00:00
sirius0xdev
2207a8a5c7
model swith
2026-04-21 02:54:01 +00:00
sirius0xdev
ed9112eb19
fix image pprocessor error
2026-04-21 02:45:41 +00:00
sirius0xdev
d78090b69c
fix quantization error
2026-04-21 02:39:30 +00:00
sirius0xdev
956ae90794
change model
2026-04-21 02:35:17 +00:00
sirius0xdev
852c481a3d
tool_call_parser error
2026-04-21 02:09:31 +00:00
sirius0xdev
d5bb38ab47
fix version problem with vllm
2026-04-21 02:00:24 +00:00
sirius0xdev
25fc06accf
fix vllm error
2026-04-21 01:49:00 +00:00
sirius0xdev
d109eba378
Update kustomization.yaml
2026-04-20 20:23:05 -04:00
Sirius Claw
29f57b48a0
Fix A100 crashloop by cleaning up invalid vllm args
2026-04-20 23:45:43 +00:00
Sirius Claw
e1cf661692
Switch L4 vLLM to Dolphin Qwen2 7B AWQ for speed and uncensored compliance
2026-04-20 20:47:30 +00:00
sirius0xdev
2f163fab36
Merge pull request #46 from sirius0xdev/feature/maximize-l4-context
...
Maximize L4 vLLM context window
2026-04-20 12:29:03 -04:00
Sirius Claw
fb951600d7
chore: maximize vllm max_model_len on L4 to 32768
2026-04-20 16:14:27 +00:00
Sirius Claw
687d61f626
fix: pin vLLM to v0.9.1 and use pythonic tool call parser
2026-04-20 07:23:43 +00:00
Sirius Claw
baa88975b5
fix: revert vLLM image to latest to support qwen3_coder tool parser
2026-04-20 07:16:42 +00:00
Sirius Claw
35dc7f01ac
fix: bump pinned vLLM version to v0.7.3 to support tool calling arguments
2026-04-20 06:58:13 +00:00
