feat(nvcf-cli): forward routing expressions unchanged

ai-tooling/user/skills/nvcf-self-managed-cli/SKILL.md

@@ -184,7 +184,7 @@ LLM function type is independent of workload packaging. For a Helm-chart backed
 
 Invocation uses the LLM route, for example `https://llm.invocation.<domain>/v1/chat/completions`. The OpenAI `model` value must be `<function-id>/<model-name>`; the function ID is the routing key and the model name is forwarded upstream.
 
-Update mutable per-model routing settings with `nvcf-cli function update --llm-model-update='name=<model>,routingMethod=<method>,tokenRateLimit=<limit>'`, or put the same fields under `modelUpdates[].llmConfig` in an update JSON file. See [reference/flags.md](reference/flags.md) for accepted routing methods. `tokenRateLimit` supports positive integer limits for `S`, `M`, `H`, `D`, and `W`; use JSON input for combined limits such as `1000-S,5000-M,100000-H,500000-D,1000000-W`. Do not include `uris` in model updates.
+Update mutable per-model routing settings with `nvcf-cli function update --llm-model-update='name=<model>,routingMethod=<method>,tokenRateLimit=<limit>'`, or put the same fields under `modelUpdates[].llmConfig` in an update JSON file. See [reference/flags.md](reference/flags.md) for the routing method format. `tokenRateLimit` supports positive integer limits for `S`, `M`, `H`, `D`, and `W`; use JSON input for combined limits such as `1000-S,5000-M,100000-H,500000-D,1000000-W`. Do not include `uris` in model updates.
 
 For `/v1/responses`, the gateway proxies the native Responses path upstream, relays SSE to streaming clients, and aggregates the terminal JSON response for non-streaming clients. For `/v1/embeddings`, input may be a string or string array, must be non-empty, and may contain at most 2048 entries.
 

ai-tooling/user/skills/nvcf-self-managed-cli/reference/commands.md

@@ -40,7 +40,7 @@ Full subcommand list. Always pair with [flags.md](flags.md) for global flags and
 | `function list` / `function list-ids` | List functions / IDs only | |
 | `function get --function-id=ID --version-id=VID` | Function metadata | |
 | `function update --function-id=ID --version-id=VID --tags=TAG[,TAG]` | Update function tags | |
-| `function update --function-id=ID --version-id=VID --llm-model-update=SPEC` | Update LLM model routing config | `SPEC` uses `name=<model>,routingMethod=<method>,tokenRateLimit=<limit>`; accepted methods are listed in [flags.md](flags.md) |
+| `function update --function-id=ID --version-id=VID --llm-model-update=SPEC` | Update LLM model routing config | `SPEC` uses `name=<model>,routingMethod=<method>,tokenRateLimit=<limit>`; the routing method format is in [flags.md](flags.md) |
 | `function deploy create --input-file=FILE` | Schedule a deployment | Blocks until ACTIVE (default 900s) |
 | `function deploy get --function-id=ID --version-id=VID [--json]` | Deployment status | `functionStatus: ACTIVE\|DEPLOYING\|ERROR\|FAILED` |
 | `function deploy update --input-file=FILE` | Modify a deployment in place | |

ai-tooling/user/skills/nvcf-self-managed-cli/reference/flags.md

@@ -81,11 +81,11 @@ Extended duration units: `s` (seconds), `m` (minutes), `h` (hours), `d` (days),
 | `--helm-chart CHART` | Helm chart URL or OCI reference. Can be used with `--function-type=LLM` for chart-packaged LLM workloads. | - |
 | `--helm-chart-service NAME` | Kubernetes Service name exposed by the chart. Required when `--helm-chart` is set. | - |
 | `--models NAME:VERSION:URI` | Standard model artifact; repeatable | - |
-| `--llm-model SPEC` | LLM model config; format `name=<model>,uris=<uri>\|<uri>,routingMethod=<method>,tokenRateLimit=<limit>`; repeatable. Token limits use `<value>-<unit>` with `S`, `M`, `H`, `D`, or `W`, for example `1000-S`. Use input JSON for combined token limits because inline specs use commas as field separators. | - |
+| `--llm-model SPEC` | LLM model config; format `name=<model>,uris=<uri>\|<uri>,routingMethod=<method>[;<param>=<value>...],tokenRateLimit=<limit>`; repeatable. Token limits use `<value>-<unit>` with `S`, `M`, `H`, `D`, or `W`, for example `1000-S`. Use input JSON for combined token limits because inline specs use commas as field separators. | - |
 | `--llm-default-priority PRIORITY` | Function-level default request priority. Lower values have higher priority. | - |
 | `--llm-per-account-priority NCA-ID:PRIORITY` | Per-account priority override; repeatable; supports up to 64 distinct NCA ID overrides. Requires a default priority. | - |
 
-In JSON and inline specs, LLM functions set `functionType: "LLM"` and model routing metadata under `models[].llmConfig`. `llmConfig.uris` declares the OpenAI-compatible upstream paths exposed by the model. Current supported paths are `/v1/chat/completions`, `/v1/responses`, and `/v1/embeddings`. `llmConfig.routingMethod` accepts `round_robin`, `power_of_two`, `groq_multiregion`, `pulsar`, or `random`.
+In JSON and inline specs, LLM functions set `functionType: "LLM"` and model routing metadata under `models[].llmConfig`. `llmConfig.uris` declares the OpenAI-compatible upstream paths exposed by the model. Current supported paths are `/v1/chat/completions`, `/v1/responses`, and `/v1/embeddings`. `llmConfig.routingMethod` is a routing algorithm name, optionally followed by `;name=value` tuning parameters, such as `pulsar;seed=stable-a`. The CLI sends it unchanged; the API rejects a malformed value and the CLI prints that error. Quote inline specs that contain `;`.
 `llmConfig.tokenRateLimit` accepts one or more comma-separated positive integer token limits in `<value>-<unit>` format. Supported units are `S` (seconds), `M` (minutes), `H` (hours), `D` (days), and `W` (weeks). Use distinct units when combining limits, for example `1000-S,5000-M,100000-H,500000-D,1000000-W` in input JSON.
 
 Helm chart packaging is independent of `functionType`. For a Helm-chart backed LLM function, set `functionType: "LLM"`, `helmChart`, `helmChartServiceName`, and `models[].llmConfig` in the same create request. `inferencePort` must be the Kubernetes Service port exposed by `helmChartServiceName`.
@@ -97,7 +97,7 @@ LLM invocation requests use `model: "<function-id>/<model-name>"`. The function
 | Flag | Purpose | Default |
 |---|---|---|
 | `--tags TAG[,TAG]` | Replace function tags | - |
-| `--llm-model-update SPEC` | LLM model update; format `name=<model>,routingMethod=<method>,tokenRateLimit=<limit>`; repeatable. Routing methods match `--llm-model`. Token limit example: `1000-S`. Use input JSON for combined token limits. | - |
+| `--llm-model-update SPEC` | LLM model update; format `name=<model>,routingMethod=<method>[;<param>=<value>...],tokenRateLimit=<limit>`; repeatable. Routing methods match `--llm-model`. Token limit example: `1000-S`. Use input JSON for combined token limits. | - |
 | `--llm-default-priority PRIORITY` | Replace the function-level priority configuration with this default and any supplied per-account overrides. | - |
 | `--llm-per-account-priority NCA-ID:PRIORITY` | Per-account priority override; repeatable; supports up to 64 distinct NCA ID overrides. Requires a default priority. | - |
 

docs/overview/cli.md

@@ -749,7 +749,7 @@ All `function create` flags:
 | `--secrets` | Secrets in `name=value` format (repeatable) |
 | `--tags` | Comma-separated tags |
 | `--models` | Model artifacts in `name:version:uri` format (repeatable) |
-| `--llm-model` | LLM model config in `name=MODEL,uris=URI\|URI,routingMethod=round_robin\|power_of_two\|groq_multiregion\|pulsar\|random,tokenRateLimit=LIMIT` format (repeatable). Token limits use `<value>-<unit>` with `S`, `M`, `H`, `D`, or `W`, for example `1000-S`. Use JSON input for combined token limits because inline model specs use commas as field separators. |
+| `--llm-model` | LLM model config in `name=MODEL,uris=URI\|URI,routingMethod=METHOD[;PARAM=VALUE...],tokenRateLimit=LIMIT` format (repeatable). The CLI forwards `routingMethod` unchanged; the API checks it. Token limits use `<value>-<unit>` with `S`, `M`, `H`, `D`, or `W`, for example `1000-S`. Use JSON input for combined token limits because inline model specs use commas as field separators. |
 | `--llm-default-priority` | Function-level default request priority. Lower values have higher priority, and `0` is highest. |
 | `--llm-per-account-priority` | Per-account override in `<nca-id>:<priority>` format. Repeatable; supports up to 64 distinct NCA ID overrides. Requires a default priority. |
 | `--resources` | Resource artifacts in `name:version:uri` format (repeatable) |
@@ -799,7 +799,7 @@ LLM functions use `functionType: "LLM"` and define model routing metadata under
 }
 ```
 
-For LLM models, `llmConfig.routingMethod` accepts `round_robin`, `power_of_two`, `groq_multiregion`, `pulsar`, or `random`.
+For LLM models, `llmConfig.routingMethod` is a routing algorithm name, optionally followed by tuning parameters, such as `pulsar;seed=stable-a`. The CLI forwards the value unchanged and prints the API error for a malformed value. For algorithms, parameters, and errors, see [LLM Request Router Load Balancing](../self-managed/llm-request-router-load-balancing.md#tune-a-model-with-a-routing-expression).
 Supported LLM paths are `/v1/chat/completions`, `/v1/responses`, and `/v1/embeddings`.
 `llmConfig.tokenRateLimit` accepts one or more comma-separated positive integer token limits in `<value>-<unit>` format. Supported units are `S` (seconds), `M` (minutes), `H` (hours), `D` (days), and `W` (weeks). Use `1000-S` for a single limit, or `1000-S,5000-M,100000-H,500000-D,1000000-W` for a combined limit with distinct units. Use JSON input for combined limits because inline CLI model specs use commas as field separators.
 

docs/overview/llm-gateway.md

@@ -92,10 +92,13 @@ gateway does not wrap it in a second NVCF envelope.
 | `/v1/embeddings` | Supports embeddings requests with string or string array input. |
 | `/v1/messages` | Preserves native Anthropic Messages JSON responses and server-sent events, including tool and thinking content blocks. |
 
-`nvcf-cli` accepts `round_robin`, `power_of_two`, `groq_multiregion`,
-`pulsar`, or `random` for `llmConfig.routingMethod`.
+`llmConfig.routingMethod` is a routing algorithm name, optionally followed by
+tuning parameters, such as `pulsar;seed=stable-a;consider_kv_free_tokens=true`.
+`nvcf-cli` forwards the value unchanged. The function API checks its format,
+and the request router checks the algorithm and parameters on each request.
 
-For the mapping to Stargate algorithms and the request-router allowlist, see
+For the mapping to Stargate algorithms, the request-router allowlist, and the
+tuning parameters, see
 [LLM Request Router Load Balancing](../self-managed/llm-request-router-load-balancing.md).
 
 `llmConfig.tokenRateLimit` applies a per-model token limit. Use one or more comma-separated limits in `<value>-<unit>` format, where `<value>` is a positive integer and `<unit>` is one of `S` (seconds), `M` (minutes), `H` (hours), `D` (days), or `W` (weeks). A single limit is one token budget over one time window, such as `1000-S`. A combined limit is multiple token budgets over distinct time windows, such as `1000-S,5000-M,100000-H,500000-D,1000000-W`; do not repeat a unit in the same value.