 Alexander Korolev
|
ae1285dfdf
fix: reduce gpu-memory-utilization to 0.75 for deepseek-v4-flash
|
il y a 3 mois |
|
Alexander Korolev
|
04e04b4e8c
fix: add initial boot delay and retry logic to auto-power-on
|
il y a 3 mois |
|
Alexander Korolev
|
09ac2dc3c4
fix(deepseek-v4-flash): revert wrong tp pp changes
|
il y a 4 mois |
|
Alexander Korolev
|
ef03b8c784
feat: auto-scale PP when pipeline_parallel_size == required_nodes
|
il y a 4 mois |
|
Alexander Korolev
|
c62934d8c3
chore: cleanup
|
il y a 4 mois |
|
Alexander Korolev
|
227eb18915
feat: auto-set VLLM_LOGGING_COLOR=1 on head node (override via recipe [env])
|
il y a 4 mois |
|
Alexander Korolev
|
338828e55d
feat: force colored vLLM logs via VLLM_LOGGING_COLOR=1
|
il y a 4 mois |
|
Alexander Korolev
|
2cf821840c
fix(deepseek): comment out shm_size (redundant with --privileged)
|
il y a 4 mois |
|
Alexander Korolev
|
7b01e19ec2
fix: HF_HOME -> /root/.cache/huggingface (fixes stale cache)
|
il y a 4 mois |
|
Alexander Korolev
|
bc41cdf4c9
fix(deepseek): restore --reasoning-config and reasoning_effort
|
il y a 4 mois |
|
Alexander Korolev
|
d9c3f845c2
fix(deepseek): align parameters with working system
|
il y a 4 mois |
|
Alexander Korolev
|
8df3f532b7
fix(deepseek): restore clean literal strings for JSON configs
|
il y a 4 mois |
|
Alexander Korolev
|
b0c139ea80
feat: auto-select smallest cluster matching required_nodes
|
il y a 4 mois |
|
Alexander Korolev
|
a9a35440a0
chore: cleanup recipes
|
il y a 4 mois |
|
Alexander Korolev
|
f5f4a2ae73
docs(deepseek): use TOML literal strings for JSON configs
|
il y a 4 mois |
|
Alexander Korolev
|
d521cfd724
docs(deepseek): use triple-quoted strings for JSON configs
|
il y a 4 mois |
|
Alexander Korolev
|
2af1104e7f
feat: launcher auto-adds --host/--port; recipes no longer hardcode
|
il y a 4 mois |
|
Alexander Korolev
|
7ba894be67
feat: per-recipe shm_size + DeepSeek V4 Flash recipe
|
il y a 4 mois |
