|
|
@@ -1898,7 +1898,8 @@ void llama_model::load_hparams(llama_model_loader & ml) {
|
|
|
ml.get_key(LLM_KV_ATTENTION_LAYERNORM_RMS_EPS, hparams.f_norm_rms_eps);
|
|
|
|
|
|
switch (hparams.n_embd) {
|
|
|
- case 1536: type = LLM_TYPE_7B_A1B; break;
|
|
|
+ case 768: type = LLM_TYPE_350M; break;
|
|
|
+ case 1536: type = (hparams.n_embd == 2048 ? LLM_TYPE_7B_A1B : LLM_TYPE_1B); break;
|
|
|
case 2048: case 2560: type = LLM_TYPE_3B; break;
|
|
|
case 4096: type = LLM_TYPE_32B; break;
|
|
|
default: type = LLM_TYPE_UNKNOWN;
|