* add auto triggered acc test * use llama 7b instead * fix env * debug download * fix download prefix * add cut dirs * fix env of model path * fix dataset download * full job * source xpu env vars * use matrix to trigger model run * reset batch=1 * remove redirect * remove some trigger * add task matrix * add precision list * test llama-7b-chat * use /mnt/disk1 to store model and datasets * remove installation test * correct downloading path * fix HF vars * add bigdl-llm env vars * rename file * fix hf_home * fix script path * rename as harness evalution * rerun  | 
			||
|---|---|---|
| .. | ||
| all-in-one | ||
| harness | ||
| pipelines | ||
| benchmark_util.py | ||
| README.md | ||
| run-benchmark-tests.sh | ||
Benchmark tool for transformers int4 (separate 1st token and rest)
benchmark_util.py is used to provide a simple benchmark tool for transformer int4 model to calculate 1st token performance and the rest on CPU and GPU.
CPU Usage
Just put this file into your benchmark directory, and then wrap your transformer int4 model with BenchmarkWrapper (model = BenchmarkWrapper(model)).
Take chatglm-6b as an example:
import torch
from bigdl.llm.transformers import AutoModel
from transformers import AutoTokenizer
from benchmark_util import BenchmarkWrapper
model_path ='THUDM/chatglm-6b'
model = AutoModel.from_pretrained(model_path, trust_remote_code=True, load_in_4bit=True)
model = BenchmarkWrapper(model, do_print=True)
tokenizer = AutoTokenizer.from_pretrained(model_path, trust_remote_code=True)
prompt = "今天睡不着怎么办"
 
with torch.inference_mode():
    input_ids = tokenizer.encode(prompt, return_tensors="pt")
    output = model.generate(input_ids, do_sample=False, max_new_tokens=32)
    output_str = tokenizer.decode(output[0], skip_special_tokens=True)
Output will be like:
=========First token cost xx.xxxxs=========
=========Last token cost average xx.xxxxs (31 tokens in all)=========
GPU Usage
Just put this file into your benchmark directory, and then wrap your transformer int4 model with BenchmarkWrapper (model = BenchmarkWrapper(model)).
Take chatglm-6b as an example:
import torch
import intel_extension_for_pytorch as ipex
from bigdl.llm.transformers import AutoModel
from transformers import AutoTokenizer
from benchmark_util import BenchmarkWrapper
model_path ='THUDM/chatglm-6b'
model = AutoModel.from_pretrained(model_path, trust_remote_code=True, load_in_4bit=True)
model = model.to('xpu')
model = BenchmarkWrapper(model, do_print=True)
tokenizer = AutoTokenizer.from_pretrained(model_path, trust_remote_code=True)
prompt = "今天睡不着怎么办"
 
with torch.inference_mode():
    # wamup two times as use ipex
    for i in range(2):
        input_ids = tokenizer.encode(prompt, return_tensors="pt").to('xpu')
        output = model.generate(input_ids, do_sample=False, max_new_tokens=32)
        output_str = tokenizer.decode(output[0], skip_special_tokens=True)
    # collect performance data now
    for i in range(5):
        input_ids = tokenizer.encode(prompt, return_tensors="pt").to('xpu')
        output = model.generate(input_ids, do_sample=False, max_new_tokens=32)
        output_str = tokenizer.decode(output[0], skip_special_tokens=True)
Output will be like:
=========First token cost xx.xxxxs=========
=========Last token cost average xx.xxxxs (31 tokens in all)=========