To use 🤗 text-generation-inference on Habana Gaudi/Gaudi2, follow these steps:
-
Build the Docker image located in this folder with:
docker build -t tgi_gaudi .
-
Launch a local server instance on 1 Gaudi card:
model=meta-llama/Llama-2-7b-hf volume=$PWD/data # share a volume with the Docker container to avoid downloading weights every run docker run -p 8080:80 -v $volume:/data --runtime=habana -e HABANA_VISIBLE_DEVICES=all -e OMPI_MCA_btl_vader_single_copy_mechanism=none --cap-add=sys_nice --ipc=host tgi_gaudi --model-id $model
-
Launch a local server instance on 8 Gaudi cards:
model=meta-llama/Llama-2-70b-hf volume=$PWD/data # share a volume with the Docker container to avoid downloading weights every run docker run -p 8080:80 -v $volume:/data --runtime=habana -e PT_HPU_ENABLE_LAZY_COLLECTIVES=true -e HABANA_VISIBLE_DEVICES=all -e OMPI_MCA_btl_vader_single_copy_mechanism=none --cap-add=sys_nice --ipc=host tgi_gaudi --model-id $model --sharded true --num-shard 8
Set
LIMIT_HPU_GRAPH=True
for larger sequence/decoding lengths(e.g. 300/212). -
You can then send a request:
curl 127.0.0.1:8080/generate \ -X POST \ -d '{"inputs":"What is Deep Learning?","parameters":{"max_new_tokens":17, "do_sample": true}}' \ -H 'Content-Type: application/json'
The first call will be slower as the model is compiled.
-
To run benchmark test, please refer TGI's benchmark tool.
To run it on the same machine, you can do the following:
docker exec -it <docker name> bash
, pick the docker started from step 3 or 4 using docker pstext-generation-benchmark -t <model-id>
, pass the model-id from docker run command- after the completion of tests, hit ctrl+c to see the performance data summary.
For gated models such as StarCoder, you will have to pass
-e HUGGING_FACE_HUB_TOKEN=<token>
to thedocker run
command above with a valid Hugging Face Hub read token.
For more information and documentation about Text Generation Inference, checkout the README of the original repo.
Not all features of TGI are currently supported as this is still a work in progress.
New changes are added for the current release:
- Sharded feature with support for DeepSpeed-inference auto tensor parallelism. Also, use HPU graphs for performance improvement.
- Torch profile.
- Batch size bucketing for decode and prefill.
Environment Variables Added:
Name | Value(s) | Default | Description | Usage |
---|---|---|---|---|
MAX_TOTAL_TOKENS | integer | 0 | Control the padding of input | add -e in docker run, such |
ENABLE_HPU_GRAPH | true/false | true | Enable hpu graph or not | add -e in docker run command |
PROF_WARMUPSTEP | integer | 0 | Enable/disable profile, control profile warmup step, 0 means disable profile | add -e in docker run command |
PROF_STEP | interger | 5 | Control profile step | add -e in docker run command |
PROF_PATH | string | /root/text-generation-inference | Define profile folder | add -e in docker run command |
LIMIT_HPU_GRAPH | True/False | False | Skip HPU graph usage for prefill to save memory, set to True for large sequence/decoding lengths(e.g. 300/212) |
add -e in docker run command |
BATCH_BUCKET_SIZE | integer | 8 | Batch size for decode operation will be rounded to the nearest multiple of this number. This limits the number of cached graphs | add -e in docker run command |
PREFILL_BATCH_BUCKET_SIZE | integer | 4 | Batch size for prefill operation will be rounded to the nearest multiple of this number. This limits the number of cached graphs | add -e in docker run command |
SKIP_TOKENIZER_IN_TGI | True/False | False | Skip tokenizer for input/output processing | add -e in docker run command |
TGI_PROFILER_ENABLED | True/False | False | Collect high-level server tracing events | add -e in docker run command |
Maximum batch size is controlled by two arguments:
- For prefill operation, please set
--max-prefill-total-tokens
asbs * max-input-length
, wherebs
is your expected maximum prefill batch size. - For decode operation, please set
--max-batch-total-tokens
asbs * max-total-tokens
, wherebs
is your expected maximum decode batch size. - Please note that batch size will be always padded to the nearest multiplication of
BATCH_BUCKET_SIZE
andPREFILL_BATCH_BUCKET_SIZE
.
The license to use TGI on Habana Gaudi is the one of TGI: https://github.com/huggingface/text-generation-inference/blob/main/LICENSE
Please reach out to [email protected] if you have any question.