Open RevanthRameshkumar opened 11 months ago
for llama model .eval(), I am getting llm.logits probs that are greater than 0...which is wrong since they are supposed to be logprobs
00000 = {float} 1.6911218166351318 00001 = {float} -0.3545471429824829 00002 = {float} -0.2773338556289673 00003 = {float} 0.3111882507801056 00004 = {float} 2.7992610931396484 00005 = {float} 3.917383909225464 00006 = {float} 3.2658145427703857 00007 = {float} 3.7564470767974854 00008 = {float} 2.6616108417510986 00009 = {float} 1.892128586769104 00010 = {float} 2.7094860076904297 00011 = {float} 2.2569401264190674 ...
for llama model .eval(), I am getting llm.logits probs that are greater than 0...which is wrong since they are supposed to be logprobs