Duplicate Models from Model Specs

[rwmehta]

Having some trouble correctly configuring librechat.yaml. What I currently have is making models show up twice, once because they are in endpoints and once from model specs. I don't understand how to just make the model specs ones show up.

version: 1.2.5
cache: true
interface:
  agents: false
  endpointsMenu: false
  modelSelect: true
  parameters: true
  presets: true
  sidePanel: true
  prompts: true
  bookmarks: true
  multiConvo: true
registration:
  socialLogins: ["google"]
modelSpecs:
  enforce: true
  prioritize: false
  list:
    - name: "gpt-4o"
      label: "Azure"
      showIconInMenu: true
      showIconInHeader: true
      default: false
      preset:
        endpoint: "azureOpenAI"
        model: "gpt-4o"
        maxContextTokens: 128000
        max_tokens: 4096
        modelLabel: "GPT 4o"
    - name: "MyCM-Genie-PT"
      label: "RAG API"
      showIconInMenu: true
      showIconInHeader: true
      default: true
      preset:
        endpoint: "RAG API"
        model: "MyCM-Genie-PT"
        maxContextTokens: 128000
        max_tokens: 4096
        modelLabel: "MyCM-Genie-PT"
        greeting: |
          greeting message 
endpoints:
  azureOpenAI:
    titleModel: "gpt-4o"
    titleConvo: true
    summarize: true
    summaryModel: "gpt-4o"
    groups:
    - group: "serverless-example"
      apiKey: "${AZUREFOUNDRY_KEY}"  
      instanceName: "name"
      version: "2024-12-01-preview" 
      models:
        gpt-4o:
          deploymentName: gpt-4o
          version: "2024-12-01-preview"
  custom:
    - name: "RAG API"
      apiKey: "api_key"
      baseURL: "url"
      models:
        default: ["MyCM-Genie-PT"]
      titleConvo: false
      summarize: false
      summaryModel: "current_model"
      forcePrompt: false
      userIdQuery: true
      addParams: 
        hello: "here"
      dropParams: ["stream"]
  agents:
    disableBuilder: true

This is how the model selection dropdown looks:

[onestardao]

What you’re seeing is caused by a config overlap — the same model is being declared in both the endpoints section and inside modelSpecs. LibreChat doesn’t automatically deduplicate them, so they each show up as separate entries.

Two ways people usually resolve this:

Keep the definition in modelSpecs only, and remove the redundant one from endpoints.

Or, if you need both sections for other providers, make sure the titleModel / modelMap fields don’t collide (so the UI can distinguish them cleanly).

In our internal checks this usually falls under a No 1.0 category (config ambiguity) — basically a small semantic drift between two sources of truth. A quick way to test is to strip down to a minimal TXTOS or wfgy core (wfgy 2.0) config, ask the model for completions, and verify only a single instance appears. If that works, then it’s just a matter of pruning or aligning your yaml.

Would you like me to sketch a minimal example yaml that avoids duplication?

[rwmehta]

An example yaml would be great. I didn't understand how to just keep the model specs definition, since it seemed like apiKey, instanceName, and baseURL only exist for endpoints not model specs.

[onestardao]

short reply — diagnostic + fixes

diagnosis
looks like a config overlap: the same model is declared twice (endpoints + modelSpecs). librechat then shows each entry separately, so you see duplicates. this is a classic config ambiguity, listed in our ProblemMap as No.10 (config overlap / model spec drift).

quick fixes you can try now

1. prefer single source of truth — keep the model entry only in modelSpecs, remove the redundant endpoint entry. that usually fixes the UI duplication immediately.
2. or make the two entries distinguishable — give different title/modelMap fields so the UI can tell them apart. this is useful if you need multiple endpoints but want them listed separately.
3. sanity test — strip to a minimal yaml with only one model, reload, confirm single entry, then reintroduce other blocks one by one to see which field collides.

if you want the formal checklist and a few small YAML examples, see ProblemMap No.10 here
https://github.com/onestardao/WFGY/blob/main/ProblemMap/README.md

if you want i can sketch a tiny example yaml that keeps the model only in modelSpecs, paste it here and i’ll adapt it to your snippet.