@@ -211,6 +211,9 @@ def run(
211211 temp : Annotated [Optional [List [float ]], typer .Option (help = "Set of temperatures to test" )] = None ,
212212 top_p : Annotated [Optional [float ], typer .Option (help = "top_p" )] = None ,
213213 top_k : Annotated [Optional [int ], typer .Option (help = "top_k" )] = None ,
214+ ctk : Annotated [Optional [str ], typer .Option (help = "ctk" )] = None ,
215+ ctv : Annotated [Optional [str ], typer .Option (help = "ctv" )] = None ,
216+ fa : Annotated [Optional [bool ], typer .Option (help = "fa" )] = None ,
214217 seed : Annotated [Optional [int ], typer .Option (help = "Random seed" )] = None ,
215218 port : Annotated [int , typer .Option (help = "llama-server port" )] = 8084 ,
216219 force : Annotated [bool , typer .Option (help = "Force overwrite of output file" )] = False ,
@@ -284,6 +287,9 @@ def elapsed():
284287 temp = t ,
285288 top_p = top_p ,
286289 top_k = top_k ,
290+ ctk = ctk ,
291+ ctv = ctv ,
292+ seed = seed ,
287293 success_ratio = float (success_count ) / n ,
288294 avg_time = mean (success_times + failure_times ),
289295 median_time = median (success_times + failure_times ),
@@ -307,6 +313,9 @@ def elapsed():
307313 server .n_ctx = n_ctx
308314 server .n_slots = 1
309315 server .jinja = True
316+ server .ctk = ctk
317+ server .ctv = ctv
318+ server .fa = fa
310319 server .n_predict = n_predict
311320 server .model_hf_repo = hf
312321 server .model_hf_file = None
0 commit comments