bug: Server doesn't forward the full header to requests

[surak]

Describe the bug

vllm provides much more info about a model than vllm-router

To Reproduce

query a model directly on vllm/v1/models and via the router

Expected behavior

this is vllm

{
    "object": "list",
    "data": [
        {
            "id": "999 NVIDIA-Nemotron-3-Nano-30B-A3B-BF16",
            "object": "model",
            "created": 1774275193,
            "owned_by": "vllm",
            "root": "models/NVIDIA-Nemotron-3-Nano-30B-A3B-BF16",
            "parent": null,
            "max_model_len": 262144,
            "permission": [
                {
                    "id": "modelperm-b5553c75cf65382c",
                    "object": "model_permission",
                    "created": 1774275193,
                    "allow_create_engine": false,
                    "allow_sampling": true,
                    "allow_logprobs": true,
                    "allow_search_indices": false,
                    "allow_view": true,
                    "allow_fine_tuning": false,
                    "organization": "*",
                    "group": null,
                    "is_blocking": false
                }

and this is vllm-router:

    {
      "id": "999 NVIDIA-Nemotron-3-Nano-30B-A3B-BF16",
      "object": "model",
      "created": 1772547439,
      "owned_by": "vllm",
      "client_count": 1
    },

Additional context

Things like max_model_len and the allow* fields are important for the clients to self adapt to a model.

[ruizhang0101]

Thanks for raising this! I think it will be worth a PR as a new feature :)) Would you want to take this one?

[yzhan1]

Hi @surak feel free to try out #927 and see if it fixes it.

cc @ruizhang0101