2bf8adc6913f17dfe8324b1b1a63df4d2b92b9a1,scripts/language_model/large_word_language_model.py,,test,#Any#Any#Any#,268

Before Change

        if max_nbatch_eval and nbatch > max_nbatch_eval:
            print("Quit evaluation early at batch %d"%nbatch)
            break
    return avg

def evaluate():
     Evaluate loop for the trained model 
    print(eval_model)

After Change

    ntotal = 0
    nbatch = 0
    hidden = eval_model.begin_state(batch_size=batch_size, func=mx.nd.zeros, ctx=ctx)
    start_time = time.time()
    for data, target, mask in data_stream:
        data = data.as_in_context(ctx)
        target = target.as_in_context(ctx)
        mask = mask.as_in_context(ctx)
        output, hidden = eval_model(data, hidden)
        hidden = detach(hidden)
        output = output.reshape((-3, -1))
        L = loss(output, target.reshape(-1,)) * mask.reshape((-1,))
        total_L += L.mean()
        ntotal += mask.mean()
        nbatch += 1
        avg = total_L / ntotal
        if nbatch % args.log_interval == 0:
            avg_scalar = float(avg.asscalar())
            ppl = math.exp(avg_scalar)
            throughput = batch_size*args.log_interval/(time.time()-start_time)
            print("Evaluation batch %d: test loss %.2f, test ppl %.2f, "
                  "throughput = %.2f samples/s"%(nbatch, avg_scalar, ppl, throughput))
            start_time = time.time()
        if max_nbatch_eval and nbatch > max_nbatch_eval:

Instances