nit, DeepSeek V1 MoE is 16B

CISC · web-flow · commit 8c41736ecc66 · 2025-03-30T11:14:44.000+02:00
diff --git a/src/llama-model.cpp b/src/llama-model.cpp
@@ -1101,7 +1101,7 @@ void llama_model::load_hparams(llama_model_loader & ml) {
                 ml.get_key(LLM_KV_EXPERT_WEIGHTS_SCALE,        hparams.expert_weights_scale);
 
                 switch (hparams.n_layer) {
-                    case 28: type = LLM_TYPE_20B; break;
+                    case 28: type = LLM_TYPE_16B; break;
                     default: type = LLM_TYPE_UNKNOWN;
                 }
             } break;

Original file line number	Diff line number	Diff line change
`@@ -1101,7 +1101,7 @@ void llama_model::load_hparams(llama_model_loader & ml) {`
`1101`	`1101`	`ml.get_key(LLM_KV_EXPERT_WEIGHTS_SCALE, hparams.expert_weights_scale);`
`1102`	`1102`
`1103`	`1103`	`switch (hparams.n_layer) {`
`1104`		`- case 28: type = LLM_TYPE_20B; break;`
	`1104`	`+ case 28: type = LLM_TYPE_16B; break;`
`1105`	`1105`	`default: type = LLM_TYPE_UNKNOWN;`
`1106`	`1106`	`}`
`1107`	`1107`	`} break;`