Make llama.cpp use tfs parameter

docs/Generation-parameters.md

@@ -10,10 +10,21 @@ llama.cpp only uses the following parameters:
 * top_p
 * top_k
 * repetition_penalty
+* tfs
 * mirostat_mode
 * mirostat_tau
 * mirostat_eta
 
+### ExLlama
+
+ExLlama only uses the following parameters:
+
+* temperature
+* top_p
+* top_k
+* repetition_penalty
+* typical_p
+
 ### RWKV
 
 RWKV only uses the following parameters when loaded through the old .pth weights:

modules/llamacpp_model.py

@@ -68,6 +68,7 @@ class LlamaCppModel:
             top_p=state['top_p'],
             top_k=state['top_k'],
             repeat_penalty=state['repetition_penalty'],
+            tfs_z=state['tfs'],
             mirostat_mode=int(state['mirostat_mode']),
             mirostat_tau=state['mirostat_tau'],
             mirostat_eta=state['mirostat_eta'],