Alexander Korolev
|
08d1e9fe03
feat(serve): add --mirror — serve from a per-node local HF mirror
|
há 2 meses atrás |
Alexander Korolev
|
4eff3e41b0
docs(skills): add nvidia-forum-reading skill
|
há 2 meses atrás |
Alexander Korolev
|
a432ba42be
docs(skills): add sm121-launcher and vllm-mod-patches skills
|
há 2 meses atrás |
Alexander Korolev
|
070bf6666a
feat(deepseek-v4-eugr): optional reasoning_content key patch (default-off)
|
há 2 meses atrás |
Alexander Korolev
|
0c7b29fbd6
feat(deepseek-v4-eugr): minimal reasoning_effort low/high/max patch (not full rewrite)
|
há 2 meses atrás |
Alexander Korolev
|
25b18fcd56
fix(deepseek-v4): drafter b12x WO routing for real DSpark acceptance + GMU 0.78
|
há 2 meses atrás |
Alexander Korolev
|
e296685e8c
fix(deepseek-v4): let DSpark survive mixed-context batches (tool-eval-bench)
|
há 2 meses atrás |
Alexander Korolev
|
5de97ee3aa
fix(deepseek-v4): get DeepSeek-V4-Flash DSpark booting on production-3.8 (ds4)
|
há 2 meses atrás |
Alexander Korolev
|
d2026c2ddf
feat(deepseek-v4): raise max-model-len to 1M + pin KV cache bytes (eugr)
|
há 2 meses atrás |
Alexander Korolev
|
3aa9d380b2
feat(recipes): agent-serving profile + GMU/env fixes from benchmark findings
|
há 2 meses atrás |
Alexander Korolev
|
0362145ef5
docs(specs): investigation — production-3.8 image update failures
|
há 2 meses atrás |
Alexander Korolev
|
a8e4932fc4
feat(deepseek-v4): add eugr b12x recipe variant with baked draft-loader mod
|
há 2 meses atrás |
Alexander Korolev
|
5bc841f211
tmp: rename to simplify
|
há 2 meses atrás |
Alexander Korolev
|
3a783ee58e
Revert "feat(deepseek-v4): move :fixed overlay to production-3.8 base + enable DSPARK tiny decode"
|
há 2 meses atrás |
Alexander Korolev
|
72f40b3da2
feat(deepseek-v4): move :fixed overlay to production-3.8 base + enable DSPARK tiny decode
|
há 2 meses atrás |
Alexander Korolev
|
e6f70ecef1
fix(deepseek-v4): forbid echoing injected content to close TC-34
|
há 2 meses atrás |
Alexander Korolev
|
8269eeafca
fix(deepseek-v4): honor thinking_mode kwarg; harden tool template vs prompt injection
|
há 2 meses atrás |
Alexander Korolev
|
bd292d7508
fix(deepseek-v4): use canonical model-card dsv4 encoding over stale bundled copy
|
há 2 meses atrás |
Alexander Korolev
|
a83c61cd7e
fix(deepseek-v4): patch vendored dsv4 tokenizer reasoning_effort handling
|
há 2 meses atrás |
Alexander Korolev
|
1bae27bd44
fix(docker): complete DS4 src runtime lock with flashinfer transitive deps
|
há 2 meses atrás |
Alexander Korolev
|
5a6ec550d7
fix(ssh): pass through already single-quoted tokens in shq
|
há 2 meses atrás |
Alexander Korolev
|
0a21797aeb
refactor(docker): extract model path-reference resolver to path_ref submodule
|
há 2 meses atrás |
Alexander Korolev
|
3e508cc2c8
feat(recipe): DSpark k=5 fix + experimental from-source DS4 build
|
há 2 meses atrás |
Alexander Korolev
|
e300668348
feat(docker): authoritative build/sync source, single-node build clusters, chunked upload
|
há 2 meses atrás |
Alexander Korolev
|
2df0e1168f
chore: remove piolium audit output and gitignore it
|
há 2 meses atrás |
Alexander Korolev
|
df4149920b
fix: remediate F1 (image-sync host-key verification) and F2 (docker run quoting)
|
há 2 meses atrás |
Alexander Korolev
|
0181645b9e
fix: make IB config optional for single-node clusters
|
há 2 meses atrás |
Alexander Korolev
|
a3bd00eef8
fix: tmux serve quoting, venv setup, cluster probe on dgx3.lan
|
há 2 meses atrás |
Alexander Korolev
|
1b4d07309a
fix: make NCCL config optional for single-node, capture stderr, support --on for tmux
|
há 2 meses atrás |
Alexander Korolev
|
c2956d7e7f
fix(i18n): use proper Fluent plural selectors for tmux-health-all-stopped
|
há 2 meses atrás |