Add missing tfs_z paramter

Author: abetlen

llama_cpp/llama.py

@@ -437,6 +437,7 @@ def generate(
         reset: bool = True,
         frequency_penalty: float = 0.0,
         presence_penalty: float = 0.0,
+        tfs_z: float = 1.0,
         mirostat_mode: int = 0,
         mirostat_tau: float = 5.0,
         mirostat_eta: float = 0.1,
@@ -495,6 +496,7 @@ def generate(
                 repeat_penalty=repeat_penalty,
                 frequency_penalty=frequency_penalty,
                 presence_penalty=presence_penalty,
+                tfs_z=tfs_z,
                 mirostat_mode=mirostat_mode,
                 mirostat_tau=mirostat_tau,
                 mirostat_eta=mirostat_eta,
@@ -576,6 +578,7 @@ def _create_completion(
         repeat_penalty: float = 1.1,
         top_k: int = 40,
         stream: bool = False,
+        tfs_z: float = 1.0,
         mirostat_mode: int = 0,
         mirostat_tau: float = 5.0,
         mirostat_eta: float = 0.1,
@@ -634,6 +637,7 @@ def _create_completion(
             top_k=top_k,
             top_p=top_p,
             temp=temperature,
+            tfs_z=tfs_z,
             mirostat_mode=mirostat_mode,
             mirostat_tau=mirostat_tau,
             mirostat_eta=mirostat_eta,
@@ -818,6 +822,7 @@ def create_completion(
         repeat_penalty: float = 1.1,
         top_k: int = 40,
         stream: bool = False,
+        tfs_z: float = 1.0,
         mirostat_mode: int = 0,
         mirostat_tau: float = 5.0,
         mirostat_eta: float = 0.1,
@@ -858,6 +863,7 @@ def create_completion(
             repeat_penalty=repeat_penalty,
             top_k=top_k,
             stream=stream,
+            tfs_z=tfs_z,
             mirostat_mode=mirostat_mode,
             mirostat_tau=mirostat_tau,
             mirostat_eta=mirostat_eta,
@@ -883,6 +889,7 @@ def __call__(
         repeat_penalty: float = 1.1,
         top_k: int = 40,
         stream: bool = False,
+        tfs_z: float = 1.0,
         mirostat_mode: int = 0,
         mirostat_tau: float = 5.0,
         mirostat_eta: float = 0.1,
@@ -923,6 +930,7 @@ def __call__(
             repeat_penalty=repeat_penalty,
             top_k=top_k,
             stream=stream,
+            tfs_z=tfs_z,
             mirostat_mode=mirostat_mode,
             mirostat_tau=mirostat_tau,
             mirostat_eta=mirostat_eta,
@@ -998,6 +1006,7 @@ def create_chat_completion(
         presence_penalty: float = 0.0,
         frequency_penalty: float = 0.0,
         repeat_penalty: float = 1.1,
+        tfs_z: float = 1.0,
         mirostat_mode: int = 0,
         mirostat_tau: float = 5.0,
         mirostat_eta: float = 0.1,
@@ -1035,6 +1044,7 @@ def create_chat_completion(
             repeat_penalty=repeat_penalty,
             presence_penalty=presence_penalty,
             frequency_penalty=frequency_penalty,
+            tfs_z=tfs_z,
             mirostat_mode=mirostat_mode,
             mirostat_tau=mirostat_tau,
             mirostat_eta=mirostat_eta,