Bug fixed with n_ctx=0 (abetlen#1015)

DanieleMorotti · web-flow · commit f1c631dc5337 · 2023-12-16T18:59:50.000-05:00
If the n_ctx is set to 0 the code should use the maximum context length of the selected model, but it didn't work. There was a problem with the initialization of this parameter and a related problem with 'n_batch'.
diff --git a/llama_cpp/llama.py b/llama_cpp/llama.py
@@ -923,6 +923,12 @@ def __init__(
         self._model = _LlamaModel(
             path_model=self.model_path, params=self.model_params, verbose=self.verbose
         )
+        # Set the default value for the context and correct the batch
+        if n_ctx == 0:
+            n_ctx = self._model.n_ctx_train()
+            self.n_batch = min(n_ctx, n_batch)
+            self.context_params.n_ctx = self._model.n_ctx_train()
+            self.context_params.n_batch = self.n_batch
 
         self._ctx = _LlamaContext(
             model=self._model,
