@@ -348,20 +348,20 @@ def test_weather(hf_repo: str, template_override: str | Tuple[str, str | None] |
348348
349349@pytest .mark .slow
350350@pytest .mark .parametrize ("result_override,n_predict,hf_repo,template_override" , [
351- (None , 128 , "bartowski/Phi-3.5-mini-instruct-GGUF:Q4_K_M" , "chatml" ),
352- (None , 128 , "bartowski/Qwen2.5-7B-Instruct-GGUF:Q4_K_M" , None ),
353- (None , 128 , "bartowski/Qwen2.5-7B-Instruct-GGUF:Q4_K_M" , "chatml" ),
354- (None , 128 , "bartowski/Hermes-2-Pro-Llama-3-8B-GGUF:Q4_K_M" , ("NousResearch/Hermes-2-Pro-Llama-3-8B" , "tool_use" )),
355- (None , 128 , "bartowski/Hermes-3-Llama-3.1-8B-GGUF:Q4_K_M" , ("NousResearch/Hermes-3-Llama-3.1-8B" , "tool_use" )),
356- (None , 128 , "bartowski/functionary-small-v3.2-GGUF:Q8_0" , ("meetkai/functionary-medium-v3.2" , None )),
357- (None , 128 , "bartowski/Meta-Llama-3.1-8B-Instruct-GGUF:Q4_K_M" , None ),
358- (None , 128 , "bartowski/Mistral-Nemo-Instruct-2407-GGUF:Q4_K_M" , None ),
359- ("^> 0.56$" , 128 , "bartowski/Mistral-Nemo-Instruct-2407-GGUF:Q4_K_M" , "chatml" ),
351+ (None , 128 , "bartowski/Phi-3.5-mini-instruct-GGUF:Q4_K_M" , "chatml" ),
352+ (None , 128 , "bartowski/Qwen2.5-7B-Instruct-GGUF:Q4_K_M" , None ),
353+ (None , 128 , "bartowski/Qwen2.5-7B-Instruct-GGUF:Q4_K_M" , "chatml" ),
354+ (None , 128 , "bartowski/Hermes-2-Pro-Llama-3-8B-GGUF:Q4_K_M" , ("NousResearch/Hermes-2-Pro-Llama-3-8B" , "tool_use" )),
355+ (None , 128 , "bartowski/Hermes-3-Llama-3.1-8B-GGUF:Q4_K_M" , ("NousResearch/Hermes-3-Llama-3.1-8B" , "tool_use" )),
356+ (None , 128 , "bartowski/functionary-small-v3.2-GGUF:Q8_0" , ("meetkai/functionary-medium-v3.2" , None )),
357+ (None , 128 , "bartowski/Meta-Llama-3.1-8B-Instruct-GGUF:Q4_K_M" , None ),
358+ (None , 128 , "bartowski/Mistral-Nemo-Instruct-2407-GGUF:Q4_K_M" , None ),
359+ ("^> 0.56$" , 128 , "bartowski/Mistral-Nemo-Instruct-2407-GGUF:Q4_K_M" , "chatml" ),
360+ (None , 128 , "bartowski/Phi-3.5-mini-instruct-GGUF:Q4_K_M" , None ),
360361
361362 # TODO: fix these (wrong results, either didn't respect decimal instruction or got wrong value)
362- ("[\\ s\\ S\\ r\\ n]*?\\ b0\\ .55644242476$" , 128 , "bartowski/Phi-3.5-mini-instruct-GGUF:Q4_K_M" , None ),
363- ("[\\ s\\ S\\ r\\ n]*?which equals 0\\ .5\\ ." , 8192 , "bartowski/DeepSeek-R1-Distill-Qwen-7B-GGUF:Q4_K_M" , None ),
364- ("[\\ s\\ S\\ r\\ n]*?\\ *\\ *Answer:\\ *\\ * 0\\ .25\\ b" , 8192 , "bartowski/DeepSeek-R1-Distill-Qwen-7B-GGUF:Q4_K_M" , ("llama-cpp-deepseek-r1" , None )),
363+ ("^The y-coordinate [\\ s\\ S]*?\\ *\\ *0.5\\ *\\ *" , 8192 , "bartowski/DeepSeek-R1-Distill-Qwen-7B-GGUF:Q4_K_M" , None ),
364+ ("[\\ s\\ S]*?\\ *\\ *0\\ .5\\ *\\ *" , 8192 , "bartowski/DeepSeek-R1-Distill-Qwen-7B-GGUF:Q4_K_M" , ("llama-cpp-deepseek-r1" , None )),
365365])
366366def test_calc_result (result_override : str | None , n_predict : int , hf_repo : str , template_override : str | Tuple [str , str | None ] | None ):
367367 global server
@@ -382,7 +382,7 @@ def test_calc_result(result_override: str | None, n_predict: int, hf_repo: str,
382382 res = server .make_request ("POST" , "/chat/completions" , data = {
383383 "max_tokens" : n_predict ,
384384 "messages" : [
385- {"role" : "system" , "content" : "You are a chatbot that uses tools/functions. Dont overthink things, and provide very concise answers. Do not explain your reasoning to the user. Provide any numerical values back to the user with two decimals." },
385+ {"role" : "system" , "content" : "You are a chatbot that uses tools/functions. Dont overthink things, and provide very concise answers. Do not explain your reasoning to the user. Provide any numerical values back to the user with at most two decimals." },
386386 {"role" : "user" , "content" : "What's the y coordinate of a point on the unit sphere at angle 30 degrees?" },
387387 {
388388 "role" : "assistant" ,
@@ -402,7 +402,7 @@ def test_calc_result(result_override: str | None, n_predict: int, hf_repo: str,
402402 "role" : "tool" ,
403403 "name" : "calculate" ,
404404 "content" : 0.55644242476 ,
405- "tool_call_id" : "call_6789" ,
405+ "tool_call_id" : "call_6789"
406406 }
407407 ],
408408 "tools" : [
@@ -434,19 +434,19 @@ def test_calc_result(result_override: str | None, n_predict: int, hf_repo: str,
434434 if result_override is not None :
435435 assert re .match (result_override , content ), f'Expected { result_override } , got { content } '
436436 else :
437- assert re .match ('^[\\ s\\ S\\ r \\ n ]*?The (y[ -])?coordinate [\\ s\\ S\\ r \\ n ]*?is (approximately )?0\\ .56\\ b|^0\\ .56$' , content ), \
437+ assert re .match ('^[\\ s\\ S]*?The (y[ -])?coordinate [\\ s\\ S]*?is (approximately )?0\\ .56\\ b|^0\\ .56$' , content ), \
438438 f'Expected something like "The y coordinate is 0.56.", got { content } '
439439
440440
441441@pytest .mark .slow
442442@pytest .mark .parametrize ("n_predict,reasoning_format,expect_content,expect_reasoning_content,hf_repo,template_override" , [
443- (1024 , 'deepseek' , "^The sum of 102 and 7 is 109.*" , None , "bartowski/Phi-3.5-mini-instruct-GGUF:Q4_K_M" , None ),
443+ (128 , 'deepseek' , "^The sum of 102 and 7 is 109.*" , None , "bartowski/Phi-3.5-mini-instruct-GGUF:Q4_K_M" , None ),
444444 (128 , None , "^The sum of 102 and 7 is 109.*" , None , "bartowski/Phi-3.5-mini-instruct-GGUF:Q4_K_M" , None ),
445445
446- (1024 , 'deepseek' , "To find the sum of.*" , "I need to calculate the sum of 102 and 7.*" , "bartowski/DeepSeek-R1-Distill-Qwen-7B-GGUF:Q4_K_M" , None ),
447- (1024 , 'none' , "<think>\n I need[\\ s\\ S\\ r \\ n ]*?</think>\n To find.*" , None , "bartowski/DeepSeek-R1-Distill-Qwen-7B-GGUF:Q4_K_M" , None ),
446+ (1024 , 'deepseek' , "To find the sum of.*" , "I need to calculate the sum of 102 and 7.*" , "bartowski/DeepSeek-R1-Distill-Qwen-7B-GGUF:Q4_K_M" , None ),
447+ (1024 , 'none' , "<think>\n ?I need[\\ s\\ S]*?</think>\n ?To find.*" , None , "bartowski/DeepSeek-R1-Distill-Qwen-7B-GGUF:Q4_K_M" , None ),
448448
449- (1024 , 'deepseek' , "To find the sum of.*" , "First, I need to add the tens place.*" , "bartowski/DeepSeek-R1-Distill-Qwen-7B-GGUF:Q4_K_M" , ("llama-cpp-deepseek-r1" , None )),
449+ (1024 , 'deepseek' , "To find the sum of.*" , "First, I [ \\ s \\ S]*" , "bartowski/DeepSeek-R1-Distill-Qwen-7B-GGUF:Q4_K_M" , ("llama-cpp-deepseek-r1" , None )),
450450])
451451def test_thoughts (n_predict : int , reasoning_format : Literal ['deepseek' , 'none' ] | None , expect_content : str | None , expect_reasoning_content : str | None , hf_repo : str , template_override : str | Tuple [str , str | None ] | None ):
452452 global server
0 commit comments