Text Generation
Transformers
PyTorch
Safetensors
bloom
Eval Results (legacy)
text-generation-inference
Instructions to use bigscience/bloom-3b with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Transformers
How to use bigscience/bloom-3b with Transformers:
# Use a pipeline as a high-level helper from transformers import pipeline pipe = pipeline("text-generation", model="bigscience/bloom-3b")# Load model directly from transformers import AutoTokenizer, AutoModelForCausalLM tokenizer = AutoTokenizer.from_pretrained("bigscience/bloom-3b") model = AutoModelForCausalLM.from_pretrained("bigscience/bloom-3b") - Notebooks
- Google Colab
- Kaggle
- Local Apps
- vLLM
How to use bigscience/bloom-3b with vLLM:
Install from pip and serve model
# Install vLLM from pip: pip install vllm # Start the vLLM server: vllm serve "bigscience/bloom-3b" # Call the server using curl (OpenAI-compatible API): curl -X POST "http://localhost:8000/v1/completions" \ -H "Content-Type: application/json" \ --data '{ "model": "bigscience/bloom-3b", "prompt": "Once upon a time,", "max_tokens": 512, "temperature": 0.5 }'Use Docker
docker model run hf.co/bigscience/bloom-3b
- SGLang
How to use bigscience/bloom-3b with SGLang:
Install from pip and serve model
# Install SGLang from pip: pip install sglang # Start the SGLang server: python3 -m sglang.launch_server \ --model-path "bigscience/bloom-3b" \ --host 0.0.0.0 \ --port 30000 # Call the server using curl (OpenAI-compatible API): curl -X POST "http://localhost:30000/v1/completions" \ -H "Content-Type: application/json" \ --data '{ "model": "bigscience/bloom-3b", "prompt": "Once upon a time,", "max_tokens": 512, "temperature": 0.5 }'Use Docker images
docker run --gpus all \ --shm-size 32g \ -p 30000:30000 \ -v ~/.cache/huggingface:/root/.cache/huggingface \ --env "HF_TOKEN=<secret>" \ --ipc=host \ lmsysorg/sglang:latest \ python3 -m sglang.launch_server \ --model-path "bigscience/bloom-3b" \ --host 0.0.0.0 \ --port 30000 # Call the server using curl (OpenAI-compatible API): curl -X POST "http://localhost:30000/v1/completions" \ -H "Content-Type: application/json" \ --data '{ "model": "bigscience/bloom-3b", "prompt": "Once upon a time,", "max_tokens": 512, "temperature": 0.5 }' - Docker Model Runner
How to use bigscience/bloom-3b with Docker Model Runner:
docker model run hf.co/bigscience/bloom-3b
Commit ·
1c5992f
1
Parent(s): 68331cd
Update README.md
Browse files
README.md
CHANGED
|
@@ -1662,7 +1662,9 @@ Please see [the BLOOM training README](https://github.com/bigscience-workshop/bi
|
|
| 1662 |
|
| 1663 |
* ALiBI positional encodings (see [paper](https://arxiv.org/pdf/2108.12409.pdf)), with GeLU activation functions
|
| 1664 |
|
| 1665 |
-
*
|
|
|
|
|
|
|
| 1666 |
|
| 1667 |
* 30 layers, 32 attention heads
|
| 1668 |
|
|
@@ -1705,18 +1707,7 @@ Please see [the BLOOM training README](https://github.com/bigscience-workshop/bi
|
|
| 1705 |
|
| 1706 |
#### **Training**
|
| 1707 |
|
| 1708 |
-
|
| 1709 |
-
_In progress._
|
| 1710 |
-
|
| 1711 |
-
Current training logs: [Tensorboard link](https://huggingface.co/tensorboard/bigscience/tr11-176B-ml-logs/)
|
| 1712 |
-
|
| 1713 |
-
- Checkpoint size:
|
| 1714 |
-
|
| 1715 |
-
- Bf16 weights: 329GB
|
| 1716 |
-
|
| 1717 |
-
- Full checkpoint with optimizer states: 2.3TB
|
| 1718 |
-
|
| 1719 |
-
- Training throughput: About 150 TFLOP per GPU per second
|
| 1720 |
|
| 1721 |
- Number of epochs: 1 (*current target*)
|
| 1722 |
|
|
@@ -1724,7 +1715,7 @@ Current training logs: [Tensorboard link](https://huggingface.co/tensorboard/big
|
|
| 1724 |
|
| 1725 |
- Started 11th March, 2022 11:42am PST
|
| 1726 |
|
| 1727 |
-
-
|
| 1728 |
|
| 1729 |
- Estimated cost of training: Equivalent of $2-5M in cloud computing (including preliminary experiments)
|
| 1730 |
|
|
|
|
| 1662 |
|
| 1663 |
* ALiBI positional encodings (see [paper](https://arxiv.org/pdf/2108.12409.pdf)), with GeLU activation functions
|
| 1664 |
|
| 1665 |
+
* 3,002,557,440 parameters:
|
| 1666 |
+
|
| 1667 |
+
* 642,252,800 embedding parameters
|
| 1668 |
|
| 1669 |
* 30 layers, 32 attention heads
|
| 1670 |
|
|
|
|
| 1707 |
|
| 1708 |
#### **Training**
|
| 1709 |
|
| 1710 |
+
Training logs: [Tensorboard link](https://huggingface.co/tensorboard/bigscience/tr11c-2B5-logs)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1711 |
|
| 1712 |
- Number of epochs: 1 (*current target*)
|
| 1713 |
|
|
|
|
| 1715 |
|
| 1716 |
- Started 11th March, 2022 11:42am PST
|
| 1717 |
|
| 1718 |
+
- Ended 5th July, 2022
|
| 1719 |
|
| 1720 |
- Estimated cost of training: Equivalent of $2-5M in cloud computing (including preliminary experiments)
|
| 1721 |
|