Tom Chen c5173a9726 Refuse a kubernetes command that does not say how many nodes it wants ...

Resolves op17-2 #1

Ray reads an unspecified node count as every node alive in the cluster, and the
kubernetes backend turned that same None into one Job completion. A caller that
asked for a multi-node conversion without naming a count therefore ran it on a
single node, which is exactly the case a conversion needing several nodes' gpu
memory cannot survive. A deployment has no alive-node set to stand in for the
missing count, so refuse it, and let the GLM-4.5 launcher pass the node count it
was configured with.

2026-09-03 22:32:10 +08:00

..

amd

Have every launch script talk to a backend object

2026-09-02 16:04:52 +08:00

models

fix: align MLA RoPE types with model configs (#2716)

2026-08-23 04:14:30 -07:00

tools

docs: label-only sidebar groups, explicit Overview pages, scoped link color (#2489)

2026-08-13 14:20:23 -07:00

run_deepseek_v4.py

Read the generation hardware table off the module that exports it

2026-09-02 18:44:34 +08:00

run_deepseek_v32.py

Have every launch script talk to a backend object

2026-09-02 16:04:52 +08:00

run_deepseek.py

Have every launch script talk to a backend object

2026-09-02 16:04:52 +08:00

run_gemma_4_26b_a4b.py

Have every launch script talk to a backend object

2026-09-02 16:04:52 +08:00

run_gemma_4_31b.py

Have every launch script talk to a backend object

2026-09-02 16:04:52 +08:00

run_glm5_1_744b_a40b_lora.py

Have every launch script talk to a backend object

2026-09-02 16:04:52 +08:00

run_glm5_2_744b_a40b_lora.py

Have every launch script talk to a backend object

2026-09-02 16:04:52 +08:00

run_glm5_2_744b_a40b.py

Have every launch script talk to a backend object

2026-09-02 16:04:52 +08:00

run_glm5_744b_a40b.py

Have every launch script talk to a backend object

2026-09-02 16:04:52 +08:00

run_glm45_355b_a32b_8node.py

Have every launch script talk to a backend object

2026-09-02 16:04:52 +08:00

run_glm45_355b_a32b.py

Refuse a kubernetes command that does not say how many nodes it wants

2026-09-03 22:32:10 +08:00

run_glm47_flash.py

Have every launch script talk to a backend object

2026-09-02 16:04:52 +08:00

run_gpt_oss_20b.py

Have every launch script talk to a backend object

2026-09-02 16:04:52 +08:00

run_inkling.py

Keep train a subcommand of the Inkling launcher

2026-09-02 18:44:34 +08:00

run_joy_ai_llm_flash.py

Have every launch script talk to a backend object

2026-09-02 16:04:52 +08:00

run_kimi_k2.py

Have every launch script talk to a backend object

2026-09-02 16:04:52 +08:00

run_kimi_k25.py

Have every launch script talk to a backend object

2026-09-02 16:04:52 +08:00

run_mcore_fsdp.py

Have every launch script talk to a backend object

2026-09-02 16:04:52 +08:00

run_nemotron_3_nano_4b_fsdp.py

Have every launch script talk to a backend object

2026-09-02 16:04:52 +08:00

run_nemotron_3_nano.py

Have every launch script talk to a backend object

2026-09-02 16:04:52 +08:00

run_nemotron_3_super_120b_a12b.py

Have every launch script talk to a backend object

2026-09-02 16:04:52 +08:00

run_nemotron_3_ultra_550b_a55b.py

Have every launch script talk to a backend object

2026-09-02 16:04:52 +08:00

run_qwen3_0_6b_fsdp.py

Have every launch script talk to a backend object

2026-09-02 16:04:52 +08:00

run_qwen3_4b_npu.py

Have every launch script talk to a backend object

2026-09-02 16:04:52 +08:00

run_qwen3_4b.py

Have every launch script talk to a backend object

2026-09-02 16:04:52 +08:00

run_qwen3_5_35b_a3b_lora.py

Have every launch script talk to a backend object

2026-09-02 16:04:52 +08:00

run_qwen3_5_35b_a3b_mtp.py

Have every launch script talk to a backend object

2026-09-02 16:04:52 +08:00

run_qwen3_6_35b_a3b_mtp.py

Have every launch script talk to a backend object

2026-09-02 16:04:52 +08:00

run_qwen3_30b_a3b_fsdp.py

Have every launch script talk to a backend object

2026-09-02 16:04:52 +08:00

run_qwen3_30b_a3b.py

Have every launch script talk to a backend object

2026-09-02 16:04:52 +08:00

run_qwen3_235b_a22b.py

Have every launch script talk to a backend object

2026-09-02 16:04:52 +08:00

run_qwen3_dense.py

Have every launch script talk to a backend object

2026-09-02 16:04:52 +08:00

run_qwen3_next_80b_a3b.py

Have every launch script talk to a backend object

2026-09-02 16:04:52 +08:00

run_qwen3_sft.py

Have every launch script talk to a backend object

2026-09-02 16:04:52 +08:00