 Alexander Korolev
|
bbe9605380
Plan: sm121 web dashboard (design doc)
|
2 月之前 |
|
Alexander Korolev
|
4275e1cd7c
config: global-config mgmt commands + ssh creds from cluster sources
|
2 月之前 |
|
Alexander Korolev
|
8446509baf
Global config: recipe_sources + cluster_sources
|
2 月之前 |
|
Alexander Korolev
|
48bbec1f66
serve: --clear flag to wipe host-side extra_mount folders (sudo)
|
2 月之前 |
|
Alexander Korolev
|
65f366427a
Add Qwen3.8-27B recipes + measured state doc
|
2 月之前 |
|
Alexander Korolev
|
50251da4ce
Bump GLM recipes to aidendle94 production-hybrid-1.6
|
2 月之前 |
|
Alexander Korolev
|
5a9795c612
Rebase deepseek-v4-flash :fixed overlay onto production-3.76.1
|
2 月之前 |
|
Alexander Korolev
|
08871e3dfa
Add --fit flag to serve: scale recipe to N nodes or whole cluster
|
2 月之前 |
|
Alexander Korolev
|
551ea7d8d7
feat(nccl): auto-detect NCCL_IB_GID_INDEX per node, opt-out via config pin
|
2 月之前 |
|
Alexander Korolev
|
42032fbe12
deepseek-v4-flash: Aiden base 3.75 + encoder fix baked in, 1M ctx, match 0rand compose tuning
|
2 月之前 |
|
Alexander Korolev
|
7f5a038adc
Tune deepseek-v4-flash-eugr: prefill threshold, prefix cache, reasoning_effort=max
|
2 月之前 |
|
Alexander Korolev
|
c6eda47a99
feat(deepseek-v4-flash-eugr): use high thinking by default
|
2 月之前 |
|
Alexander Korolev
|
43c9bd346f
recipe(deepseek-v4-flash-eugr): second-variant test config
|
2 月之前 |
|
Alexander Korolev
|
5d90e025f0
revert: keep experimental eugr recipe changes out of main (user is still testing)
|
2 月之前 |
|
Alexander Korolev
|
08d1e9fe03
feat(serve): add --mirror — serve from a per-node local HF mirror
|
2 月之前 |
|
Alexander Korolev
|
4eff3e41b0
docs(skills): add nvidia-forum-reading skill
|
2 月之前 |
|
Alexander Korolev
|
a432ba42be
docs(skills): add sm121-launcher and vllm-mod-patches skills
|
2 月之前 |
|
Alexander Korolev
|
070bf6666a
feat(deepseek-v4-eugr): optional reasoning_content key patch (default-off)
|
2 月之前 |
|
Alexander Korolev
|
0c7b29fbd6
feat(deepseek-v4-eugr): minimal reasoning_effort low/high/max patch (not full rewrite)
|
2 月之前 |
|
Alexander Korolev
|
25b18fcd56
fix(deepseek-v4): drafter b12x WO routing for real DSpark acceptance + GMU 0.78
|
2 月之前 |
|
Alexander Korolev
|
e296685e8c
fix(deepseek-v4): let DSpark survive mixed-context batches (tool-eval-bench)
|
2 月之前 |
|
Alexander Korolev
|
5de97ee3aa
fix(deepseek-v4): get DeepSeek-V4-Flash DSpark booting on production-3.8 (ds4)
|
2 月之前 |
|
Alexander Korolev
|
d2026c2ddf
feat(deepseek-v4): raise max-model-len to 1M + pin KV cache bytes (eugr)
|
2 月之前 |
|
Alexander Korolev
|
3aa9d380b2
feat(recipes): agent-serving profile + GMU/env fixes from benchmark findings
|
2 月之前 |
|
Alexander Korolev
|
0362145ef5
docs(specs): investigation — production-3.8 image update failures
|
2 月之前 |
|
Alexander Korolev
|
a8e4932fc4
feat(deepseek-v4): add eugr b12x recipe variant with baked draft-loader mod
|
2 月之前 |
|
Alexander Korolev
|
5bc841f211
tmp: rename to simplify
|
2 月之前 |
|
Alexander Korolev
|
3a783ee58e
Revert "feat(deepseek-v4): move :fixed overlay to production-3.8 base + enable DSPARK tiny decode"
|
2 月之前 |
|
Alexander Korolev
|
72f40b3da2
feat(deepseek-v4): move :fixed overlay to production-3.8 base + enable DSPARK tiny decode
|
2 月之前 |
|
Alexander Korolev
|
e6f70ecef1
fix(deepseek-v4): forbid echoing injected content to close TC-34
|
2 月之前 |
