llama3-8b-accelerator / config.json
JRosenkranz's picture
initial commit of weights
972329a
raw
history blame
No virus
326 Bytes
{
"architectures": [
"MLPSpeculatorPreTrainedModel"
],
"emb_dim": 4096,
"inner_dim": 3072,
"model_type": "mlp_speculator",
"n_candidates": 5,
"n_predict": 4,
"top_k_tokens_per_head": [
4,
3,
2,
2
],
"torch_dtype": "float16",
"transformers_version": "4.38.2",
"vocab_size": 128256
}