internlm2¶
Model type |
|
Class |
|
Task |
|
Source |
|
Description¶
InternLM2 causal language model with HF-aligned weight names.
Module attributes match HF naming to minimize preprocess_weights. Only tok_embeddings→embed_tokens, wo→o_proj renames and the grouped/interleaved wqkv split remain.
The wqkv weight has a grouped/interleaved layout: [num_kv_heads, (num_q_per_kv_group + 2), head_dim, hidden_size]
Usage¶
mobius build --model <MODEL_ID> output_dir/
from mobius import build
model = build("<MODEL_ID>")