DeepSeek-V3.1-Terminus-AWQ

#4
by panweiguo - opened

Can the DeepSeek-V3.1-Terminus-AWQ version of the model be quantized? Thank you.

QuantTrio org

The results didn’t meet our expectations — the model tends to generate incoherent outputs.
Even with the FP16Mix strategy, it still requires a precision close to 8-bit to respond properly.
Considering this, we’ve decided not to release the DeepSeek-V3.1-Terminus-AWQ.

JunHowie changed discussion status to closed

Sign up or log in to comment