vLLM Compatibility Problem with GPT OSS 120B and OpenClaw by spark-vllm-docker

Turtle7777 · February 11, 2026, 8:53am

Hi NV member:
I would like to use the script from "GitHub - eugr/spark-vllm-docker: Docker configuration for running VLLM on dual DGX Sparks
" to quickly set up a GPT OSS 120B server so that OpenClaw can access this server to run related services. However, after following the steps below, I encounter a “model not found” issue. Could this be caused by some incompatibility in vLLM?

I also noticed that GitHub - fidecastro/fix_glm46v: A fix for OpenClaw to work with GLM4.5 and GLM4.6V
addresses related issues. Could you please advise how this problem can be fixed?

Thank you.

Setup the vllm + gpt-oss-120b server by the following command and link
GitHub - eugr/spark-vllm-docker: Docker configuration for running VLLM on dual DGX Sparks

Full setup: build container + download model + run

./run-recipe.sh openai-gpt-oss-120b --solo --setup

Setup the Openclaw
a. Install and config
Install: Install - OpenClaw
Config for vllm: Local Models - OpenClaw
My config file:
openclaw.txt (2.3 KB)
Have the following error happen when chat in openclaw ui

(APIServer pid=176) WARNING 02-11 08:06:30 [protocol.py:117] The following fields were present in the request but ignored: {‘strict’}
(APIServer pid=176) WARNING 02-11 08:06:30 [protocol.py:117] The following fields were present in the request but ignored: {‘strict’}
(APIServer pid=176) WARNING 02-11 08:06:30 [protocol.py:117] The following fields were present in the request but ignored: {‘strict’}
(APIServer pid=176) WARNING 02-11 08:06:30 [protocol.py:117] The following fields were present in the request but ignored: {‘strict’}
(APIServer pid=176) WARNING 02-11 08:06:30 [protocol.py:117] The following fields were present in the request but ignored: {‘strict’}
(APIServer pid=176) WARNING 02-11 08:06:30 [protocol.py:117] The following fields were present in the request but ignored: {‘strict’}
(APIServer pid=176) WARNING 02-11 08:06:30 [protocol.py:117] The following fields were present in the request but ignored: {‘strict’}
(APIServer pid=176) WARNING 02-11 08:06:30 [protocol.py:117] The following fields were present in the request but ignored: {‘strict’}
(APIServer pid=176) WARNING 02-11 08:06:30 [protocol.py:117] The following fields were present in the request but ignored: {‘strict’}
(APIServer pid=176) WARNING 02-11 08:06:30 [protocol.py:117] The following fields were present in the request but ignored: {‘strict’}
(APIServer pid=176) WARNING 02-11 08:06:30 [protocol.py:117] The following fields were present in the request but ignored: {‘strict’}
(APIServer pid=176) WARNING 02-11 08:06:30 [protocol.py:117] The following fields were present in the request but ignored: {‘strict’}
(APIServer pid=176) WARNING 02-11 08:06:30 [protocol.py:117] The following fields were present in the request but ignored: {‘strict’}
(APIServer pid=176) WARNING 02-11 08:06:30 [protocol.py:117] The following fields were present in the request but ignored: {‘strict’}
(APIServer pid=176) WARNING 02-11 08:06:30 [protocol.py:117] The following fields were present in the request but ignored: {‘strict’}
(APIServer pid=176) WARNING 02-11 08:06:30 [protocol.py:117] The following fields were present in the request but ignored: {‘strict’}
(APIServer pid=176) WARNING 02-11 08:06:30 [protocol.py:117] The following fields were present in the request but ignored: {‘strict’}
(APIServer pid=176) WARNING 02-11 08:06:30 [protocol.py:117] The following fields were present in the request but ignored: {‘strict’}
(APIServer pid=176) WARNING 02-11 08:06:30 [protocol.py:117] The following fields were present in the request but ignored: {‘strict’}
(APIServer pid=176) WARNING 02-11 08:06:30 [protocol.py:117] The following fields were present in the request but ignored: {‘strict’}
(APIServer pid=176) WARNING 02-11 08:06:30 [protocol.py:117] The following fields were present in the request but ignored: {‘strict’}
(APIServer pid=176) WARNING 02-11 08:06:30 [protocol.py:117] The following fields were present in the request but ignored: {‘strict’}
(APIServer pid=176) WARNING 02-11 08:06:30 [protocol.py:117] The following fields were present in the request but ignored: {‘store’}
(APIServer pid=176) ERROR 02-11 08:06:30 [serving_chat.py:236] Error with model error=ErrorInfo(message=‘The model gpt-oss-120b does not exist.’, type=‘NotFoundError’, param=‘model’, code=404)
(APIServer pid=176) INFO: 127.0.0.1:47302 - “POST /v1/chat/completions HTTP/1.1” 404 Not Found

adg1 · February 11, 2026, 9:04am

From a cursory look, it seems the APIServer is getting invoked using a model id that differs from the one being served.

Guessing here: vllm is serving openai/gpt-oss-120b but you are referencing simply gpt-oss-120b. Fix: Try updating the model id in the API calls.

Turtle7777 · February 11, 2026, 9:48am

Hi @adg1
My OpenClaw configuration is as follows. How should this part be configured for the best results?

},
“models”: {
“providers”: {
“openai”: {
“baseUrl”: “http://127.0.0.1:8000/v1”,
“apiKey”: “OPENAI_KEY”,
“api”: “openai-completions”,
“models”: [
{
“id”: “openai/gpt-oss-120b”,
“name”: “openai/gpt-oss-120b”,
“reasoning”: false,
“input”: [
“text”
],
“cost”: {
“input”: 0,
“output”: 0,
“cacheRead”: 0,
“cacheWrite”: 0
},
“contextWindow”: 200000,
“maxTokens”: 8192
}
]
}
}
},
“agents”: {
“defaults”: {
“model”: {
“primary”: “openai/gpt-oss-120b”
},
“workspace”: “/home/asus/.openclaw/workspace”,
“maxConcurrent”: 4,
“subagents”: {
“maxConcurrent”: 8
}
}
},

adg1 · February 11, 2026, 11:06am

Hello!

I have no experience with OpenClaw, and I cannot be more precise, unfortunately.

That being said, the APIServer logs suggest a model Id mismatch – I may be mistaken.

Now, from your OpenClaw configuration it is not apparent the presence of wrong model identifiers. Hence I wonder whether further identifiers are present elsewhere in the system configuration or code.

I hope the above helps you dig the root cause.

Keyper-AI · February 11, 2026, 8:20pm

When launching with a ./run-recipe.sh set or via /.launch-cluster.sh set the custom arg with --served-model-name gpt-oss-120b

Then in OpenClaw config have the ID be the same “gpt-oss-120b”

    "models": [
      {
        "id": "Keyper-Thinker",
        "name": "Keyper Thinker",
        "reasoning": false,
        "input": [
          "text"
        ],
        "cost": {
          "input": 0,
          "output": 0,
          "cacheRead": 0,
          "cacheWrite": 0
        },
        "contextWindow": 165000,
        "maxTokens": 8192
      }
    ]

./launch-cluster.sh  \
exec vllm serve \
  QuantTrio/MiniMax-M2.1-AWQ \
  --served-model-name Keyper-Thinker \
  --port 8000 \
  --host 0.0.0.0 \
  --gpu-memory-utilization 0.85 \
  -tp 2 \
  --distributed-executor-backend ray \
  --max-model-len 165000 \
  --load-format fastsafetensors \
  --enable-auto-tool-choice \
  --tool-call-parser minimax_m2 \
  --reasoning-parser minimax_m2 \
  --trust-remote-code

eugr · February 11, 2026, 8:30pm

Looks like your browser autocorrected double dash to a single one, the argument is --served-model-name.

Keyper-AI · February 11, 2026, 8:31pm

Fixed. Gracias.

Turtle7777 · February 12, 2026, 2:19am

Hi @Keyper-AI and @eugr :
After adding the --served-model-name gpt-oss-120b argument and updating the OpenClaw ID, I no longer see the “gpt-oss-120b” failure. However, the vLLM server still shows a 400 Bad Request error.

I tested using Ollama with gpt-oss-120b, and it works normally. Is there any specific configuration I should double-check or pay attention to?

Thank you for your help.

Add the --served-model-name gpt-oss-120b to spark-vllm-docker

asus@gx10-9680:~/Desktop/test/eugr/spark-vllm-docker$ git diff
diff --git a/recipes/openai-gpt-oss-120b.yaml b/recipes/openai-gpt-oss-120b.yaml
index 09cfa52..88cbdd6 100644
— a/recipes/openai-gpt-oss-120b.yaml
+++ b/recipes/openai-gpt-oss-120b.yaml
@@ -36,6 +36,7 @@ command: |
vllm serve openai/gpt-oss-120b
–tool-call-parser openai
–reasoning-parser openai_gptoss \
--served-model-name gpt-oss-120b \
 --enable-auto-tool-choice \
 --tensor-parallel-size {tensor_parallel} \
 --distributed-executor-backend ray \

Run the following command to enable server
$ ./run-recipe.sh openai-gpt-oss-120b --solo --setup
Modify the openclaw.json

“models”: {
“providers”: {
“openai”: {
“baseUrl”: “http://127.0.0.1:8000/v1”,
“apiKey”: “OPENAI_KEY”,
“api”: “openai-completions”,
“models”: [
{
“id”: “gpt-oss-120b”,
“name”: “gpt oss 120b”,
“reasoning”: false,
“input”: [
“text”
],
“cost”: {
“input”: 0,
“output”: 0,
“cacheRead”: 0,
“cacheWrite”: 0
},
“contextWindow”: 200000,
“maxTokens”: 8192
}
]
}
}
},
“agents”: {
“defaults”: {
“model”: {
“primary”: “openai/gpt-oss-120b”
},
“workspace”: “/home/asus/.openclaw/workspace”,
“maxConcurrent”: 4,
“subagents”: {
“maxConcurrent”: 8
}
}
},

Fail to ask openclaw because the following error

(APIServer pid=176) WARNING 02-12 02:05:35 [protocol.py:117] The following fields were present in the request but ignored: {‘strict’}
(APIServer pid=176) WARNING 02-12 02:05:35 [protocol.py:117] The following fields were present in the request but ignored: {‘strict’}
(APIServer pid=176) WARNING 02-12 02:05:35 [protocol.py:117] The following fields were present in the request but ignored: {‘strict’}
(APIServer pid=176) WARNING 02-12 02:05:35 [protocol.py:117] The following fields were present in the request but ignored: {‘strict’}
(APIServer pid=176) WARNING 02-12 02:05:35 [protocol.py:117] The following fields were present in the request but ignored: {‘strict’}
(APIServer pid=176) WARNING 02-12 02:05:35 [protocol.py:117] The following fields were present in the request but ignored: {‘strict’}
(APIServer pid=176) WARNING 02-12 02:05:35 [protocol.py:117] The following fields were present in the request but ignored: {‘strict’}
(APIServer pid=176) WARNING 02-12 02:05:35 [protocol.py:117] The following fields were present in the request but ignored: {‘strict’}
(APIServer pid=176) WARNING 02-12 02:05:35 [protocol.py:117] The following fields were present in the request but ignored: {‘strict’}
(APIServer pid=176) WARNING 02-12 02:05:35 [protocol.py:117] The following fields were present in the request but ignored: {‘strict’}
(APIServer pid=176) WARNING 02-12 02:05:35 [protocol.py:117] The following fields were present in the request but ignored: {‘strict’}
(APIServer pid=176) WARNING 02-12 02:05:35 [protocol.py:117] The following fields were present in the request but ignored: {‘strict’}
(APIServer pid=176) WARNING 02-12 02:05:35 [protocol.py:117] The following fields were present in the request but ignored: {‘strict’}
(APIServer pid=176) WARNING 02-12 02:05:35 [protocol.py:117] The following fields were present in the request but ignored: {‘strict’}
(APIServer pid=176) WARNING 02-12 02:05:35 [protocol.py:117] The following fields were present in the request but ignored: {‘strict’}
(APIServer pid=176) WARNING 02-12 02:05:35 [protocol.py:117] The following fields were present in the request but ignored: {‘strict’}
(APIServer pid=176) WARNING 02-12 02:05:35 [protocol.py:117] The following fields were present in the request but ignored: {‘strict’}
(APIServer pid=176) WARNING 02-12 02:05:35 [protocol.py:117] The following fields were present in the request but ignored: {‘strict’}
(APIServer pid=176) WARNING 02-12 02:05:35 [protocol.py:117] The following fields were present in the request but ignored: {‘strict’}
(APIServer pid=176) WARNING 02-12 02:05:35 [protocol.py:117] The following fields were present in the request but ignored: {‘store’}
(APIServer pid=176) INFO: 127.0.0.1:60938 - “POST /v1/chat/completions HTTP/1.1” 400 Bad Request

eugr · February 12, 2026, 2:35am

You still use openai/gpt-oss-120b here while using just gpt-oss-120b earlier.
And gpt-oss-120b is a reasoning model.

Also, gpt-oss-120b doesn’t support context window over 131072 tokens.

I don’t use openclaw/clawdbot, so the problem might be elsewhere, but I’d fix these things first.

Turtle7777 · February 12, 2026, 6:43am

Hi @Keyper-AI, @eugr and @adg1
It is now working normally with OpenClaw in the Spark-VLLM-Docker + GPT-OSS-120B environment after applying the following steps and modifications. Thank you for your helpful suggestions and methods.

Setup the vllm + gpt-oss-120b server by the following command and link
GitHub - eugr/spark-vllm-docker: Docker configuration for running VLLM on dual DGX Sparks
a. Add the --served-model-name gpt-oss-120b to openai-gpt-oss-120b.yaml

diff --git a/recipes/openai-gpt-oss-120b.yaml b/recipes/openai-gpt-oss-120b.yaml
index 09cfa52..88cbdd6 100644
— a/recipes/openai-gpt-oss-120b.yaml
+++ b/recipes/openai-gpt-oss-120b.yaml
@@ -36,6 +36,7 @@ command: |
vllm serve openai/gpt-oss-120b
–tool-call-parser openai
–reasoning-parser openai_gptoss \

–served-model-name gpt-oss-120b
–enable-auto-tool-choice
–tensor-parallel-size {tensor_parallel}
–distributed-executor-backend ray \

b. Full setup: build container + download model + run

./run-recipe.sh openai-gpt-oss-120b --solo --setup

Setup the Openclaw
a. Install and config
Install: Install - OpenClaw
Config for vllm: Local Models - OpenClaw
My config file:
openclaw_vllm_gpt-oss-120b_final.txt (3.5 KB)

“models”: {
“providers”: {
“openai”: {
“baseUrl”: “http://127.0.0.1:8000/v1”,
“apiKey”: “OPENAI_KEY”,
“api”: “openai-completions”,
“models”: [
{
“id”: “gpt-oss-120b”,
“name”: “gpt-oss-120b”,
“reasoning”: false,
“input”: [
“text”
],
“cost”: {
“input”: 0,
“output”: 0,
“cacheRead”: 0,
“cacheWrite”: 0
},
“contextWindow”: 200000,
“maxTokens”: 8192
}
]
}
}
},
“agents”: {
“defaults”: {
“model”: {
“primary”: “openai/gpt-oss-120b”
},
“workspace”: “/home/asus/.openclaw/workspace”,
“compaction”: {
“mode”: “safeguard”
},
“maxConcurrent”: 4,
“subagents”: {
“maxConcurrent”: 8
}
}
},

The vllm gpt-oss-120b server can work normally

(APIServer pid=175) WARNING 02-12 06:09:14 [protocol.py:117] The following fields were present in the request but ignored: {‘strict’}
(APIServer pid=175) WARNING 02-12 06:09:14 [protocol.py:117] The following fields were present in the request but ignored: {‘strict’}
(APIServer pid=175) WARNING 02-12 06:09:14 [protocol.py:117] The following fields were present in the request but ignored: {‘strict’}
(APIServer pid=175) WARNING 02-12 06:09:14 [protocol.py:117] The following fields were present in the request but ignored: {‘strict’}
(APIServer pid=175) WARNING 02-12 06:09:14 [protocol.py:117] The following fields were present in the request but ignored: {‘strict’}
(APIServer pid=175) WARNING 02-12 06:09:14 [protocol.py:117] The following fields were present in the request but ignored: {‘strict’}
(APIServer pid=175) WARNING 02-12 06:09:14 [protocol.py:117] The following fields were present in the request but ignored: {‘strict’}
(APIServer pid=175) WARNING 02-12 06:09:14 [protocol.py:117] The following fields were present in the request but ignored: {‘strict’}
(APIServer pid=175) WARNING 02-12 06:09:14 [protocol.py:117] The following fields were present in the request but ignored: {‘strict’}
(APIServer pid=175) WARNING 02-12 06:09:14 [protocol.py:117] The following fields were present in the request but ignored: {‘strict’}
(APIServer pid=175) WARNING 02-12 06:09:14 [protocol.py:117] The following fields were present in the request but ignored: {‘strict’}
(APIServer pid=175) WARNING 02-12 06:09:14 [protocol.py:117] The following fields were present in the request but ignored: {‘strict’}
(APIServer pid=175) WARNING 02-12 06:09:14 [protocol.py:117] The following fields were present in the request but ignored: {‘strict’}
(APIServer pid=175) WARNING 02-12 06:09:14 [protocol.py:117] The following fields were present in the request but ignored: {‘strict’}
(APIServer pid=175) WARNING 02-12 06:09:14 [protocol.py:117] The following fields were present in the request but ignored: {‘strict’}
(APIServer pid=175) WARNING 02-12 06:09:14 [protocol.py:117] The following fields were present in the request but ignored: {‘strict’}
(APIServer pid=175) WARNING 02-12 06:09:14 [protocol.py:117] The following fields were present in the request but ignored: {‘strict’}
(APIServer pid=175) WARNING 02-12 06:09:14 [protocol.py:117] The following fields were present in the request but ignored: {‘strict’}
(APIServer pid=175) WARNING 02-12 06:09:14 [protocol.py:117] The following fields were present in the request but ignored: {‘strict’}
(APIServer pid=175) WARNING 02-12 06:09:14 [protocol.py:117] The following fields were present in the request but ignored: {‘store’}
(APIServer pid=175) INFO: 127.0.0.1:47234 - “POST /v1/chat/completions HTTP/1.1” 200 OK
(APIServer pid=175) INFO 02-12 06:09:16 [loggers.py:257] Engine 000: Avg prompt throughput: 2138.2 tokens/s, Avg generation throughput: 51.4 tokens/s, Running: 1 reqs, Waiting: 0 reqs, GPU KV cache usage: 2.2%, Prefix cache hit rate: 80.3%
(APIServer pid=175) INFO 02-12 06:09:26 [loggers.py:257] Engine 000: Avg prompt throughput: 0.0 tokens/s, Avg generation throughput: 11.0 tokens/s, Running: 0 reqs, Waiting: 0 reqs, GPU KV cache usage: 0.0%, Prefix cache hit rate: 80.3%
(APIServer pid=175) INFO 02-12 06:09:36 [loggers.py:257] Engine 000: Avg prompt throughput: 0.0 tokens/s, Avg generation throughput: 0.0 tokens/s, Running: 0 reqs, Waiting: 0 reqs, GPU KV cache usage: 0.0%, Prefix cache hit rate: 80.3%

eugr · February 12, 2026, 6:46am

BTW, you don’t have to modify the recipe now. You can just add extra vLLM arguments after -- in `run-recipe.sh’, like this:

./run-recipe.sh openai-gpt-oss-120b --solo --setup -- --served-model-name gpt-oss-120b

Turtle7777 · February 12, 2026, 6:49am

Got it ! I learned something new.
Thank you!

eugr · February 12, 2026, 7:11am

We added it a couple of days ago :)

Turtle7777 · February 12, 2026, 7:14am

Haha, I’m really lucky! It’s a great way to make the setup much more convenient.

Turtle7777 · February 12, 2026, 8:27am

Hi @eugr :
As you mentioned, when my chat gets longer and longer, perhaps because the context becomes larger and larger, after it is sent to the gpt-oss-120b server, the following error keeps occurring. I assume every model has this issue, right? Or based on your understanding, are there better models or approaches to handle this?
Thank you.

Error message:

400 max_tokens must be at least 1, got -17474. (parameter=max_tokens, value=-17474)

Turtle7777 · February 12, 2026, 9:12am

The contextPruning setting in OPENCLAW might be a good way to reduce the input of large amounts of contextual content.

contextPruning:
Configuration Reference - OpenClaw

eugr · February 12, 2026, 5:18pm

Don’t set “maxTokens”: 8192, it’s too low and restricts model to output anything longer than that. Just don’t set it at all. Also reduce your context window parameter in openclaw settings to 131072 - maximum that gpt-oss-120b supports.

Turtle7777 · February 13, 2026, 2:27am

Hi @eugr :
Thank you for your suggestion. You’re absolutely right. As you mentioned, the context window should better align with GPT-OSS-120B.

We can experiment later with OpenClaw plus multiple agents (Claude Code), combined with either a single or multiple DGX Spark systems. That would also be an interesting application scenario for DGX Spark edge devices.

rdariolemes · February 26, 2026, 3:08am

Hello Team,

We have a two-node cluster and have been testing several models for use with OpenClaw agents, but we haven’t had much success so far.

Could you please recommend which model has been delivering the best overall performance and user experience?

Thank you.

Turtle7777 · March 2, 2026, 6:19am

Hi @rdariolemes
Haha, I’m just a user, but you could list the setup process clearly — that might be more helpful.

Topic		Replies	Views
Total nightmare : NEMOCLAW over Paperclip over OPENCLAW over vLLM over Dokers, over LLM flavours , over Linux DGX Spark / GB10	14	2889	March 25, 2026
Has anyone had any good experience running on DGX Spark with clawdbot? DGX Spark / GB10	38	4152	March 20, 2026
Guide: Gemma 4 31B on DGX Spark via NemoClaw — Dual-Model Setup Guide DGX Spark / GB10 Projects nim , llama , nemotron , nemoclaw , openclaw	3	1311	April 10, 2026
GLM-4.7-Flash-NVFP4 was just released, but for Transformers 5.0 + vLLM 0.14...? DGX Spark / GB10	90	4221	February 27, 2026
Install and Use vLLM for Inference on two Sparks does not work DGX Spark / GB10	159	4920	December 9, 2025
Why does OPENCLAW call the OLLAMA API, and the model doesn't remember my previous question? How to set up OLLAMA? DGX Spark / GB10	8	1075	March 17, 2026
NemoClaw on Spark DGX Spark / GB10 agentic-ai	56	4676	March 24, 2026
Who wants to be the hero and help a total newbie! Got a spark and um, yeah DGX Spark / GB10 nemotron	8	491	April 17, 2026
OpenClaw w/ Nemotron-3-Super NVFP4 TensorRT inference on Spark Discussion DGX Spark / GB10 nemotron	14	1296	April 9, 2026
Moving from Mac to NVIDIA: bought powerful hardware, but drowning in configs DGX Spark / GB10 llama , nemotron	37	2222	February 25, 2026

vLLM Compatibility Problem with GPT OSS 120B and OpenClaw by spark-vllm-docker

Full setup: build container + download model + run

Related topics