llama : recognize 1B phi models (#4847)

teleprint-me · web-flow · commit 329ff615699d · 2024-01-10T15:39:09.000+02:00
This update categorizes models with 24 layers as MODEL_1B, ensuring compatibility with different Phi model variants without impacting existing Phi-2 model functionality.
diff --git a/llama.cpp b/llama.cpp
@@ -2829,6 +2829,7 @@ static void llm_load_hparams(
                 ml.get_key(LLM_KV_ATTENTION_LAYERNORM_EPS, hparams.f_norm_eps);
 
                 switch (hparams.n_layer) {
+                    case 24: model.type = e_model::MODEL_1B; break;
                     case 32: model.type = e_model::MODEL_3B; break;
                     default: model.type = e_model::MODEL_UNKNOWN;
                 }

Original file line number	Diff line number	Diff line change
`@@ -2829,6 +2829,7 @@ static void llm_load_hparams(`
`2829`	`2829`	`ml.get_key(LLM_KV_ATTENTION_LAYERNORM_EPS, hparams.f_norm_eps);`
`2830`	`2830`
`2831`	`2831`	`switch (hparams.n_layer) {`
	`2832`	`+ case 24: model.type = e_model::MODEL_1B; break;`
`2832`	`2833`	`case 32: model.type = e_model::MODEL_3B; break;`
`2833`	`2834`	`default: model.type = e_model::MODEL_UNKNOWN;`
`2834`	`2835`	`}`