2023-12-05 10:12:16 +00:00
<!-- -
Copyright 2023 The HuggingFace Team. All rights reserved.
2022-10-08 10:30:12 +00:00
2023-12-05 10:12:16 +00:00
Licensed under the Apache License, Version 2.0 (the "License");
you may not use this file except in compliance with the License.
You may obtain a copy of the License at
2023-02-03 11:43:37 +00:00
2023-12-05 10:12:16 +00:00
http://www.apache.org/licenses/LICENSE-2.0
2023-07-19 11:38:52 +00:00
2023-12-05 10:12:16 +00:00
Unless required by applicable law or agreed to in writing, software
distributed under the License is distributed on an "AS IS" BASIS,
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
See the License for the specific language governing permissions and
limitations under the License.
-->
2023-07-19 11:38:52 +00:00
2023-12-05 10:12:16 +00:00
# Text Generation Inference on Habana Gaudi
2023-02-13 12:02:45 +00:00
2023-12-05 10:12:16 +00:00
To use [🤗 text-generation-inference ](https://github.com/huggingface/text-generation-inference ) on Habana Gaudi/Gaudi2, follow these steps:
2023-07-25 17:45:25 +00:00
2023-12-05 10:12:16 +00:00
1. Build the Docker image located in this folder with:
```bash
docker build -t tgi_gaudi .
```
2. Launch a local server instance on 1 Gaudi card:
```bash
model=meta-llama/Llama-2-7b-hf
volume=$PWD/data # share a volume with the Docker container to avoid downloading weights every run
2023-07-25 17:45:25 +00:00
2023-12-05 10:12:16 +00:00
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 $model
```
2024-03-10 21:02:15 +00:00
> For gated models such as [LLama](https://huggingface.co/meta-llama) or [StarCoder](https://huggingface.co/bigcode/starcoder), you will have to pass `-e HUGGING_FACE_HUB_TOKEN=<token>` to the `docker run` command above with a valid Hugging Face Hub read token.
2023-12-05 10:12:16 +00:00
3. Launch a local server instance on 8 Gaudi cards:
```bash
model=meta-llama/Llama-2-70b-hf
volume=$PWD/data # share a volume with the Docker container to avoid downloading weights every run
2023-07-25 17:45:25 +00:00
2023-12-05 10:12:16 +00:00
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
```
2024-03-10 21:02:15 +00:00
4. You can then send a simple request:
2023-12-05 10:12:16 +00:00
```bash
curl 127.0.0.1:8080/generate \
-X POST \
2024-03-10 21:02:15 +00:00
-d '{"inputs":"What is Deep Learning?","parameters":{"max_new_tokens":32}}' \
2023-12-05 10:12:16 +00:00
-H 'Content-Type: application/json'
```
2024-03-10 21:02:15 +00:00
5. To run static benchmark test, please refer to [TGI's benchmark tool ](https://github.com/huggingface/text-generation-inference/tree/main/benchmark ).
2023-02-13 12:02:45 +00:00
2023-12-05 10:12:16 +00:00
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 ps
* `text-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.
2023-02-08 16:53:33 +00:00
2023-12-05 10:12:16 +00:00
For more information and documentation about Text Generation Inference, checkout [the README ](https://github.com/huggingface/text-generation-inference#text-generation-inference ) of the original repo.
2023-02-08 16:53:33 +00:00
2023-12-05 10:12:16 +00:00
Not all features of TGI are currently supported as this is still a work in progress.
2024-03-10 21:02:15 +00:00
TGI on Intel Gaudi has been validated mainly with Llama model. Support for other models from Optimum Habana will be added successively.
2023-02-08 16:53:33 +00:00
2024-01-11 13:51:49 +00:00
2024-03-10 21:02:15 +00:00
## Setup TGI
2023-02-08 16:53:33 +00:00
2024-03-10 21:02:15 +00:00
Maximum sequence length is controlled by two arguments:
- `--max-input-length` is the maximum possible input prompt length. Default value is `1024` .
- `--max-total-tokens` is the maximum possible total length of the sequence (input and output). Default value is `2048` .
2023-02-08 16:53:33 +00:00
2024-03-10 21:02:15 +00:00
Maximum batch size is controlled by two arguments:
- For prefill operation, please set `--max-prefill-total-tokens` as `bs * max-input-length` , where `bs` is your expected maximum prefill batch size.
- For decode operation, please set `--max-batch-total-tokens` as `bs * max-total-tokens` , where `bs` is your expected maximum decode batch size.
- Please note that batch size will be always padded to the nearest multiplication of `BATCH_BUCKET_SIZE` and `PREFILL_BATCH_BUCKET_SIZE` .
Environment variables:
2023-02-08 16:53:33 +00:00
2024-03-10 21:02:15 +00:00
< div align = "left" >
2023-07-28 07:14:03 +00:00
2024-02-23 00:52:14 +00:00
| Name | Value(s) | Default | Description | Usage |
| --------------------------- | :--------- | :--------------- | :------------------------------------------------------------------------------------------------------------------------------- | :--------------------------- |
2024-03-01 10:25:15 +00:00
| ENABLE_HPU_GRAPH | True/False | True | Enable hpu graph or not | 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 |
2024-02-23 00:52:14 +00:00
| 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 |
| PAD_SEQUENCE_TO_MULTIPLE_OF | integer | 128 | For prefill operation, sequences will be padded to a multiple of provided value. | 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 |
2024-02-28 09:40:13 +00:00
| WARMUP_ENABLED | True/False | True | Enable warmup during server initialization to recompile all graphs. This can increase TGI setup time. | add -e in docker run command |
2024-02-29 09:56:26 +00:00
| QUEUE_THRESHOLD_MS | integer | 120 | Controls the threshold beyond which the request are considered overdue and handled with priority. Shorter requests are prioritized otherwise. | add -e in docker run command |
2023-12-05 10:12:16 +00:00
< / div >
2023-07-28 07:14:03 +00:00
2024-03-10 21:02:15 +00:00
## Profiler
To collect performance profiling, please set below environment variables:
< div align = "left" >
| Name | Value(s) | Default | Description | Usage |
| ------------------ | :--------- | :--------------- | :------------------------------------------------------- | :--------------------------- |
| PROF_WAITSTEP | integer | 0 | Control profile wait steps | add -e in docker run command |
| PROF_WARMUPSTEP | integer | 0 | Control profile warmup steps | add -e in docker run command |
| PROF_STEP | integer | 0 | Enable/disable profile, control profile active steps | add -e in docker run command |
| PROF_PATH | string | /tmp/hpu_profile | Define profile folder | add -e in docker run command |
| PROF_RANKS | string | 0 | Comma-separated list of ranks to profile | add -e in docker run command |
| PROF_RECORD_SHAPES | True/False | False | Control record_shapes option in the profiler | add -e in docker run command |
< / div >
2023-08-11 14:46:08 +00:00
2024-01-11 13:51:49 +00:00
2023-12-05 10:12:16 +00:00
> 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 api-enterprise@huggingface.co if you have any question.