vocab: Use int32 for special tokens

Special tokens are currently read as uint32 from the model metadata.
However, all other parts of the system (including the tokenizer) use
int32 to represent tokens so it is impossible to represent the high
portion of the unsigned range. For consistency and to avoid casts,
we should just use int32 everywhere.

model/llama/model.go

@@ -35,8 +35,8 @@ func New(c ml.Config) (model.Model, error) {
 				Values: c.Strings("tokenizer.ggml.tokens"),
 				Types:  c.Uints("tokenizer.ggml.token_type"),
 				Merges: c.Strings("tokenizer.ggml.merges"),
-				BOS:    c.Uint("tokenizer.ggml.bos_token_id"),
-				EOS:    c.Uint("tokenizer.ggml.eos_token_id"),
+				BOS:    int32(c.Uint("tokenizer.ggml.bos_token_id")),
+				EOS:    int32(c.Uint("tokenizer.ggml.eos_token_id")),
 			},
 		),
 		Layers: make([]Layer, c.Uint("block_count")),