[BUG] Inference predictions don't match Huggingface model's #2184
Description
Describe the bug
HF output:
[{'generated_text': 'DeepSpeed is the greatest force of nature. It is not only energy which, due to how well it creates energy, is very powerful. Some may say that the universe is only one such entity; others may claim that the cosmos is all a result'}]
Deepspeed output with mp size = 1,
[{'generated_text': "DeepSpeed is the greatest speed at which your avatar is at (not how fast you can reach or fall, but how far you can fall).\n\nOn iOS and Android, if the player has reached or fallen too far they won't fly."}]
To Reproduce
modelname = "gpt2"
pipe = pipeline("text-generation", model="gpt2", device=0, framework="pt")
model = pipe.model
if dtype == "fp16":
pipe.model.half()
query_text = "DeepSpeed is the greatest"
torch.cuda.synchronize()
start = time.time()
hf_output = pipe(query_text)
torch.cuda.synchronize()
print("hf output", hf_output)
import deepspeed
pipe.model = deepspeed.init_inference(
pipe.model,
mp_size=1,
dtype=torch.half,
replace_method="auto",
replace_with_kernel_inject=True,
)
torch.cuda.synchronize()
start = time.time()
ds_output = pipe(query_text)
torch.cuda.synchronize()
ds_time = time.time() - start
print(hf_output, ds_output)
Expected behavior
Outputs expected to be the same
ds_report output
--------------------------------------------------
DeepSpeed C++/CUDA extension op report
--------------------------------------------------
NOTE: Ops not installed will be just-in-time (JIT) compiled at
runtime if needed. Op compatibility means that your system
meet the required dependencies to JIT install the op.
--------------------------------------------------
JIT compiled ops requires ninja
ninja .................. [OKAY]
--------------------------------------------------
op name ................ installed .. compatible
--------------------------------------------------
cpu_adam ............... [NO] ....... [OKAY]
cpu_adagrad ............ [NO] ....... [OKAY]
fused_adam ............. [NO] ....... [OKAY]
fused_lamb ............. [NO] ....... [OKAY]
[WARNING] please install triton==1.0.0 if you want to use sparse attention
sparse_attn ............ [NO] ....... [NO]
transformer ............ [NO] ....... [OKAY]
stochastic_transformer . [NO] ....... [OKAY]
[WARNING] async_io requires the dev libaio .so object and headers but these were not found.
[WARNING] async_io: please install the libaio-dev package with apt
[WARNING] If libaio is already installed (perhaps from source), try setting the CFLAGS and LDFLAGS environment variables to where it can be found.
async_io ............... [NO] ....... [NO]
utils .................. [NO] ....... [OKAY]
quantizer .............. [NO] ....... [OKAY]
transformer_inference .. [NO] ....... [OKAY]
--------------------------------------------------
DeepSpeed general environment info:
torch install path ............... ['/opt/conda/lib/python3.8/site-packages/torch']
torch version .................... 1.11.0+cu113
torch cuda version ............... 11.3
torch hip version ................ None
nvcc version ..................... 11.3
deepspeed install path ........... ['/opt/conda/lib/python3.8/site-packages/deepspeed']
deepspeed info ................... 0.7.0, unknown, unknown
deepspeed wheel compiled w. ...... torch 1.11, cuda 11.3
Screenshots
NA
System info (please complete the following information):
- OS: [e.g. Ubuntu 18.04] Ubuntu
- GPU count and types A100 GPU
- Interconnects (if applicable) [e.g., two machines connected with 100 Gbps IB] N/A
- Python version 3.8.3
- Any other relevant info about your setup
Launcher context
inference, single process
Docker context
Additional context
Add any other context about the problem here.