diff --git a/docs/model-workflows/shared-cache.md b/docs/model-workflows/shared-cache.md index dae262854..09fab28a1 100644 --- a/docs/model-workflows/shared-cache.md +++ b/docs/model-workflows/shared-cache.md @@ -107,7 +107,9 @@ when transferring it, not just a snapshot directory. ## 3. Validate as the consumer, then load a local snapshot On a reader node (or inside the actual workload container, with its real UID and -mount), run the validator from the repository root: +mount), use a DeepOps checkout present there with +`scripts/validation/validate_model_cache.py` readable and its parent directories +traversable by the consumer UID; run the validator from that checkout's root: ```bash python3 scripts/validation/validate_model_cache.py --json \ diff --git a/docs/model-workflows/vllm-quickstart.md b/docs/model-workflows/vllm-quickstart.md index 527d97b17..717d4f4bb 100644 --- a/docs/model-workflows/vllm-quickstart.md +++ b/docs/model-workflows/vllm-quickstart.md @@ -39,8 +39,11 @@ is needed. Review the model license before use. production isolation, multi-node serving, and scheduler manifests are outside this example. The validator deliberately has no authentication options. -The commands below run **on the allocated serving node**, from the repository -root. Package installation can download large GPU dependencies. For disconnected +The commands below run **on the allocated serving node**, from a DeepOps checkout +present there with `scripts/validation/validate_model_cache.py` and +`scripts/validation/validate_vllm.py` readable and their parent directories +traversable by the consumer UID; run from that checkout's root. +Package installation can download large GPU dependencies. For disconnected systems, prepare an approved wheel mirror first; offline model loading does not make package installation offline. diff --git a/skills/manage-model-cache/SKILL.md b/skills/manage-model-cache/SKILL.md index e68429426..95dfbe40f 100644 --- a/skills/manage-model-cache/SKILL.md +++ b/skills/manage-model-cache/SKILL.md @@ -34,8 +34,11 @@ for the supported CLI version, exact configuration and working download example. as shown in the guide. Include every tokenizer/config file and weight shard required by the workload. Do not use `main`, `--local-dir`, or deprecated `hf_transfer`; do not have compute jobs warm the cache themselves. -3. Stop writes to the selected snapshot. As the actual reader UID, in the actual - workload mount context, run `python3 scripts/validation/validate_model_cache.py` +3. Stop writes to the selected snapshot. Use a DeepOps checkout on the reader node + (or inside the workload container) with `scripts/validation/validate_model_cache.py` + readable and its parent directories traversable by the consumer UID. + From that checkout's root, as the actual reader UID in the actual workload + mount context, run `python3 scripts/validation/validate_model_cache.py` with `--json`, `--cache-dir`, `--repo-id`, `--revision`, and one `--require-file` per needed file. The guide has a complete command for the public example. 4. Accept only exit `0` **and** `ok: true`. Record revision, `checked_files` and diff --git a/skills/serve-model/SKILL.md b/skills/serve-model/SKILL.md index 7a750b7aa..03e1a0cc4 100644 --- a/skills/serve-model/SKILL.md +++ b/skills/serve-model/SKILL.md @@ -5,7 +5,10 @@ description: Use when serving a cached model on one GPU with vLLM, checking a lo # Serve a model on one node -Run from the repository root on the allocated serving node. Read the +Use a DeepOps checkout on the allocated serving node (or inside the workload +container) with `scripts/validation/validate_model_cache.py` and +`scripts/validation/validate_vllm.py` readable and their parent directories +traversable by the consumer UID; run from that checkout's root. Read the [vLLM quickstart](../../docs/model-workflows/vllm-quickstart.md) for the exact installation and foreground launch commands. This skill does not allocate a GPU or authorize a workload.