I'm not sure if this is an LM Studio issue, Codex, or Model but here goes anyway...
Which version of LM Studio?
LM Studio 0.4.19
Which operating system?
LM Studio on Windows 11
Codex CLI on CentOS 10
What is the bug?
Stream disconnected before completion: Engine protocol predict stream returned an error: {"code":500,"message":"The model produced output that does not match the expected peg-native format","type":"server_error"}
Screenshots
root@dev:/opt/nvc-web# oss
╭───────────────────────────────────────────────────────────╮
│ >_ OpenAI Codex (v0.144.6) │
│ │
│ model: openai/gpt-oss-20b medium /model to change │
│ directory: /opt/nvc-web │
│ permissions: YOLO mode │
╰───────────────────────────────────────────────────────────╯
Tip: New Build faster with Codex.
• Goal active Objective: in the front end when displaying the reports Alternative Responses Formal, Informal, etc... the 'cells/cards' are all 5 on one line, and i want them to be each on their own line
⚠ Model metadata for openai/gpt-oss-20b not found. Defaulting to fallback metadata; this can degrade performance and cause issues.
• Explored
└ List ls -R
Read report-render.mjs
Search report-alt-grid in styles
Read report.css, responsive.css
• Reconnecting... 2/5 (2m 35s • esc to interrupt) · 3 background terminals running · /ps to view · /stop to close
└ Stream disconnected before completion: Engine protocol predict stream returned an error: {"code":500,"message":"The model produced output that does not match the expected peg-native format","type":"server_error"}
› Use /skills to list available skills
openai/gpt-oss-20b medium · /opt/nvc-web
Logs
codex-bug.txt
server-bug.txt
To Reproduce
Steps to reproduce the behavior:
Unable to reproduce on-demand. Happens randomly 3-4 times a day for me during normal working.
I'm not sure if this is an LM Studio issue, Codex, or Model but here goes anyway...
Which version of LM Studio?
LM Studio 0.4.19
Which operating system?
LM Studio on Windows 11
Codex CLI on CentOS 10
What is the bug?
Stream disconnected before completion: Engine protocol predict stream returned an error: {"code":500,"message":"The model produced output that does not match the expected peg-native format","type":"server_error"}
Screenshots
root@dev:/opt/nvc-web# oss
╭───────────────────────────────────────────────────────────╮
│ >_ OpenAI Codex (v0.144.6) │
│ │
│ model: openai/gpt-oss-20b medium /model to change │
│ directory: /opt/nvc-web │
│ permissions: YOLO mode │
╰───────────────────────────────────────────────────────────╯
Tip: New Build faster with Codex.
• Goal active Objective: in the front end when displaying the reports Alternative Responses Formal, Informal, etc... the 'cells/cards' are all 5 on one line, and i want them to be each on their own line
⚠ Model metadata for
openai/gpt-oss-20bnot found. Defaulting to fallback metadata; this can degrade performance and cause issues.• Explored
└ List ls -R
Read report-render.mjs
Search report-alt-grid in styles
Read report.css, responsive.css
• Reconnecting... 2/5 (2m 35s • esc to interrupt) · 3 background terminals running · /ps to view · /stop to close
└ Stream disconnected before completion: Engine protocol predict stream returned an error: {"code":500,"message":"The model produced output that does not match the expected peg-native format","type":"server_error"}
› Use /skills to list available skills
openai/gpt-oss-20b medium · /opt/nvc-web
Logs
codex-bug.txt
server-bug.txt
To Reproduce
Steps to reproduce the behavior:
Unable to reproduce on-demand. Happens randomly 3-4 times a day for me during normal working.