search

songmzhang / DSKD

Repo for Paper "Dual-Space Knowledge Distillation for Large Language Models".
29 stars 3 forks source link

Evaluation script error with TinyLlama #12

Closed srikhetramohanty closed 1 week ago

srikhetramohanty commented 1 month ago

Hi @songmzhang I tried running the evaluation shell script with TinyLlama and it threw the error :

image

Any leads into what might be the issue here? Mistral checkpoint ran fine with this modified evaluation shell script.

Note : MODEL_PATH="/---/---/DSKD/model_hub/tinyllama/tinyllama-1.1b-3T"

Terminal command : bash scripts/tinyllama/eval_sample.sh $”/---/---/DSKD/outputs/tinyllama/tinyllama-1.1b-3T/sft/criterion=cross_entropy__lora-rank=256-alpha=8-dropout=0.1-bf16__epoch=10__bsz=8x1x1=8__lr=0.001/epoch10_step14300_loss2.2730_rougel27.6927” $16

songmzhang commented 1 month ago

Hi, this is because we missed adding the tokenizer class of TinyLLama to the def get_tokenizer() in utils.py. Now we have fixed this bug and you can pull the latest code version.

srikhetramohanty commented 3 weeks ago

Thanks. It worked.