mlc-llm
mlc-llm copied to clipboard
[Model Request] phi-4-mini-instruct
⚙️ Request New Models
- Link to an existing implementation (e.g. Hugging Face/Github): (https://huggingface.co/microsoft/Phi-4-mini-instruct)
- Is this model architecture supported by MLC-LLM? (the list of supported models)
Additional context
A few hours ago, phi-4-mini-instruct was released. I encountered the following error after converting it using convert_weight and running the code.
org.apache.tvm.Base$TVMError: TVMError: Assert fail: T.Cast("int32", phi3_fused_rope_longrope_scaling_ext_factors_handle_shape[0]) == 64, Argument phi3_fused_rope_longrope_scaling.ext_factors_handle.shape[0] has an unsatisfied constraint: 64 == T.Cast("int32", phi3_q4f16_0_18ed5b1a1cf36b050b3ce55681639a49_fused_rope_longrope_scaling_ext_factors_handle_shape[0])
I set model_type to phi-3, but based on the error, it seems to be caused by the fact that phi-3 does not have a process for applying rotary_dim. Is there a way to convert it properly?