CHECKPOINT_PATH=conver-4tp-model

VOCAB_FILE=gpt2-vocab.json
MERGE_FILE=gpt2-merges.txt
GPUS_PER_NODE=4
NNODES=1

DISTRIBUTED_ARGS="--nproc_per_node $GPUS_PER_NODE --nnodes $NNODES "
python -m torch.distributed.launch $DISTRIBUTED_ARGS tools/generate_samples_gpt.py \
       --tensor-model-parallel-size 4 \
       --num-layers 24 \
       --hidden-size 1024 \
       --load $CHECKPOINT_PATH \
       --num-attention-heads 16 \
       --max-position-embeddings 1024 \
       --tokenizer-type GPT2BPETokenizer \
       --fp16 \
       --micro-batch-size 2 \
       --seq-length 1024 \
       --out-seq-length 128 \
       --temperature 1.0 \
       --vocab-file $VOCAB_FILE \
       --merge-file $MERGE_FILE \
       --genfile gpt2-genfile.json \
       --num-samples 4 \
       --top_p 0.9 \
       --recompute \