Context too large for some models?

[JulienMaille]

Trying to redirect to GLM-5.2 and getting error message below

  "_default_provider": "nvidia",
  "_default_model": "z-ai/glm-5.2",
  "_provider_models": {
    "default": {
      "nvidia": "z-ai/glm-5.2",
      "zen": "deepseek-v4-flash-free"
    },
    "gemini-3.5-flash": {
      "nvidia": "z-ai/glm-5.2"
    },

[nvidia] API error 400: {"message":"This model's maximum context length is 202752 tokens. However, your messages resulted in 231205 tokens. Please reduce the length of the messages.","type":"Bad Request","code":400}

Is that expected? Should we add automated context truncation?

[12errh]

@JulienMaille

Two options:

1. Set the contex to lite in config tab it will reduce system promt 66%.
2. if possibel move to new session. because if you keep chating in 1 session the contex get biger and biger.
3. use different model. i only use nvidia setp 3.7 flash as defult model and for my main model i always use mimo 2.5 or deep seek using opencode zen provider. it work perfectly
4. if you can not move you session. then change a thing just add opencode zen api and use mimo 2.5 or deep seek v4 flash. or even nemotron 3 ultra it have 1m contex window.

NOTE: i start work on it automatic contex compaction when reaches to 90 percent or 80 config able.

[JulienMaille]

1. Lite mode already set
2. That was medium size session
   3./4. Well no, the whole point was to use free frontier GML-5.2 which support 1.000.000 token context: https://docs.api.nvidia.com/nim/reference/z-ai-glm-5.2

[12errh]

@JulienMaille

1. Nvidia hosted glm 5.2 have only 200k contex window not 1m. so this is there problme.
2. i am working on contex compaction and that per provider model fallback feture. so saty tuned

[JulienMaille]

How do you check the context size provided by Nvidia for that model?

[12errh]

@JulienMaille

I checked the GLM-5.2 model card, and it states that the model supports a 1M-token context window. However, I couldn't find NVIDIA explicitly stating that their hosted deployment exposes the full 1M context. From my testing, the NVIDIA API reports a 200K maximum context window, and requests beyond that limit fail.

It's possible that NVIDIA's hosted deployment currently serves a 200K-context configuration rather than the full 1M capability.

[12errh]

@JulienMaille
i make an simple discord server. what about talking on discord beaceu here chatting is very difficult (https://discord.gg/Zye9Gkwys)