Description
The framework adds an empty content message along with the actual response into the content array of the [
{ "type": "text", "text": "" }, // ← empty first element
{ "type": "text", "text": "previous assistant response" } // ← actual response
]?
Due to this, I believe the chat template parser malfunctions and forces the model to respond to my previous and current message together
I observed this behaviour in GPTOSS 120B on VLLM
Code Sample
Error Messages / Stack Traces
Package Versions
1.5.0
.NET Version
No response
Additional Context
No response
Description
The framework adds an empty content message along with the actual response into the content array of the [
{ "type": "text", "text": "" }, // ← empty first element
{ "type": "text", "text": "previous assistant response" } // ← actual response
]?
Due to this, I believe the chat template parser malfunctions and forces the model to respond to my previous and current message together
I observed this behaviour in GPTOSS 120B on VLLM
Code Sample
Error Messages / Stack Traces
Package Versions
1.5.0
.NET Version
No response
Additional Context
No response