Fix for the --batch-size parameter (rustformers#94)

* Added support to cached sessions in REPL mode

* Fixed the --batch-size parameter, the default value was being used regardless of the value provided by the user.

llama-rs/src/lib.rs

@@ -1461,7 +1461,7 @@ impl InferenceSession {
             return Err(InferenceError::ContextFull);
         }
 
-        for batch in prompt_tokens.chunks(8) {
+        for batch in prompt_tokens.chunks(params.n_batch) {
             model.evaluate(self, params, batch, &mut EvaluateOutputRequest::default());
             for &tk in batch {
                 // NOTE: No string ever tokenizes to the end of sentence. So we