evolla¶
Model type |
|
Class |
|
Task |
|
Source |
|
Description¶
Standard causal language model with TextModel backbone and LM head.
Compatible with Llama 2/3, Mistral, Qwen2/2.5, and other architectures that follow the standard decoder-only transformer pattern with GQA and RoPE.
Replicates HuggingFace’s LlamaForCausalLM, MistralForCausalLM,
Qwen2ForCausalLM, etc.
Inputs: input_ids, attention_mask, position_ids, past_key_values. Outputs: logits (batch, seq_len, vocab_size), present_key_values.
Usage¶
mobius build --model <MODEL_ID> output_dir/
from mobius import build
model = build("<MODEL_ID>")