Skip to content

armgit / txtai

Public Imported from GitHub
0

💡 All-in-one AI framework for semantic search, LLM orchestration and language model workflows

Add generator pipeline for text generation, closes #416

davidmezzetti0bfb5961ce04
diff --git a/docs/pipeline/index.md b/docs/pipeline/index.md
index 91b61392..9f6f7986 100644
--- a/docs/pipeline/index.md
+++ b/docs/pipeline/index.md
@@ -25,6 +25,7 @@ The following is a list of the default pipelines available in txtai.
- Text
- [Entity](text/entity)
- [Extractive QA](text/extractor)
+ - [Generator](text/generator)
- [Labeling](text/labels)
- [Sequences](text/sequences)
- [Similarity](text/similarity)
diff --git a/docs/pipeline/text/generator.md b/docs/pipeline/text/generator.md
new file mode 100644
index 00000000..bae01e3c
--- /dev/null
+++ b/docs/pipeline/text/generator.md
@@ -0,0 +1,62 @@
+# Generator
+
+![pipeline](../../images/pipeline.png#only-light)
+![pipeline](../../images/pipeline-dark.png#only-dark)
+
+The Generator pipeline takes an input prompt and generates follow-on text.
+
+## Example
+
+The following shows a simple example using this pipeline.
+
+```python
+from txtai.pipeline import Generator
+
+# Create and run pipeline
+generator = Generator()
+generator("Hello, how are you?")
+```
+
+## Configuration-driven example
+
+Pipelines are run with Python or configuration. Pipelines can be instantiated in [configuration](../../../api/configuration/#pipeline) using the lower case name of the pipeline. Configuration-driven pipelines are run with [workflows](../../../workflow/#configuration-driven-example) or the [API](../../../api#local-instance).
+
+### config.yml
+```yaml
+# Create pipeline using lower case class name
+generator:
+
+# Run pipeline with workflow
+workflow:
+ generator:
+ tasks:
+ - action: generator
+```
+
+### Run with Workflows
+
+```python
+from txtai.app import Application
+
+# Create and run pipeline with workflow
+app = Application("config.yml")
+list(app.workflow("generator", ["Hello, how are you?"]))
+```
+
+### Run with API
+
+```bash
+CONFIG=config.yml uvicorn "txtai.api:app" &
+
+curl \
+ -X POST "http://localhost:8000/workflow" \
+ -H "Content-Type: application/json" \
+ -d '{"name":"generator", "elements": ["Hello, how are you?"]}'
+```
+
+## Methods
+
+Python documentation for the pipeline.
+
+### ::: txtai.pipeline.Generator.__init__
+### ::: txtai.pipeline.Generator.__call__
diff --git a/mkdocs.yml b/mkdocs.yml
index 1332a376..2bc4cb9a 100644
--- a/mkdocs.yml
+++ b/mkdocs.yml
@@ -73,6 +73,7 @@ nav:
- Text:
- Entity: pipeline/text/entity.md
- Extractor: pipeline/text/extractor.md
+ - Generator: pipeline/text/generator.md
- Labels: pipeline/text/labels.md
- Sequences: pipeline/text/sequences.md
- Similarity: pipeline/text/similarity.md
diff --git a/src/python/txtai/pipeline/text/__init__.py b/src/python/txtai/pipeline/text/__init__.py
index 261439d4..615fd598 100644
--- a/src/python/txtai/pipeline/text/__init__.py
+++ b/src/python/txtai/pipeline/text/__init__.py
@@ -5,6 +5,7 @@ Text imports
from .crossencoder import CrossEncoder
from .entity import Entity
from .extractor import Extractor
+from .generator import Generator
from .labels import Labels
from .questions import Questions
from .sequences import Sequences
diff --git a/src/python/txtai/pipeline/text/generator.py b/src/python/txtai/pipeline/text/generator.py
new file mode 100644
index 00000000..bd4dc80b
--- /dev/null
+++ b/src/python/txtai/pipeline/text/generator.py
@@ -0,0 +1,72 @@
+"""
+Generator module
+"""
+
+from ..hfpipeline import HFPipeline
+
+
+class Generator(HFPipeline):
+ """
+ Generate text with a causal language model.
+ """
+
+ def __init__(self, path=None, quantize=False, gpu=True, model=None):
+ super().__init__(self.task(), path, quantize, gpu, model)
+
+ def __call__(self, text, prefix=None, maxlength=512, workers=0):
+ """
+ Generates text using input text
+
+ Args:
+ text: text|list
+ prefix: optional prefix to prepend to text elements
+ maxlength: maximum sequence length
+ workers: number of concurrent workers to use for processing data, defaults to None
+
+ Returns:
+ generated text
+ """
+
+ # List of texts
+ texts = text if isinstance(text, list) else [text]
+
+ # Add prefix, if necessary
+ if prefix:
+ texts = [f"{prefix}{x}" for x in texts]
+
+ # Run pipeline
+ results = self.pipeline(texts, max_length=maxlength, num_workers=workers)
+
+ # Get generated text
+ results = [self.clean(x) for x in results]
+
+ return results[0] if isinstance(text, str) else results
+
+ def clean(self, result):
+ """
+ Applies a series of rules to clean generated text.
+
+ Args:
+ result: input result
+
+ Returns:
+ clean text
+ """
+
+ # Extract output from list, if necessary
+ result = result[0] if isinstance(result, list) else result
+
+ # Get generated text field
+ text = result["generated_text"]
+
+ return text.replace("$=", "<=")
+
+ def task(self):
+ """
+ Get the pipeline task name.
+
+ Returns:
+ pipeline task name
+ """
+
+ return "text-generation"
diff --git a/src/python/txtai/pipeline/text/sequences.py b/src/python/txtai/pipeline/text/sequences.py
index c45b131c..3c64d87e 100644
--- a/src/python/txtai/pipeline/text/sequences.py
+++ b/src/python/txtai/pipeline/text/sequences.py
@@ -2,55 +2,13 @@
Sequences module
"""
-from ..hfpipeline import HFPipeline
+from .generator import Generator
-class Sequences(HFPipeline):
+class Sequences(Generator):
"""
Runs text through a sequence-sequence model.
"""
- def __init__(self, path=None, quantize=False, gpu=True, model=None):
- super().__init__("text2text-generation", path, quantize, gpu, model)
-
- def __call__(self, text, prefix=None, maxlength=512, workers=0):
- """
- Runs a sequence-sequence model for input texts.
-
- Args:
- text: text|list
- prefix: optional prefix to prepend to text elements
- maxlength: maximum sequence length
- workers: number of concurrent workers to use for processing data, defaults to None
-
- Returns:
- generated text
- """
-
- # List of texts
- texts = text if isinstance(text, list) else [text]
-
- # Add prefix, if necessary
- if prefix:
- texts = [f"{prefix}{x}" for x in texts]
-
- # Run text2text pipeline
- results = self.pipeline(texts, max_length=maxlength, num_workers=workers)
-
- # Get generated text
- results = [self.clean(x["generated_text"]) for x in results]
-
- return results[0] if isinstance(text, str) else results
-
- def clean(self, text):
- """
- Applies a series of rules to clean generated text.
-
- Args:
- text: input text
-
- Returns:
- clean text
- """
-
- return text.replace("$=", "<=")
+ def task(self):
+ return "text2text-generation"
diff --git a/test/python/testpipeline/testgenerator.py b/test/python/testpipeline/testgenerator.py
new file mode 100644
index 00000000..be126ca3
--- /dev/null
+++ b/test/python/testpipeline/testgenerator.py
@@ -0,0 +1,24 @@
+"""
+Generator module tests
+"""
+
+import unittest
+
+from txtai.pipeline import Generator
+
+
+class TestGenerator(unittest.TestCase):
+ """
+ Sequences tests.
+ """
+
+ def testGeneration(self):
+ """
+ Test text pipeline generation
+ """
+
+ model = Generator("hf-internal-testing/tiny-random-gpt2")
+ start = "Hello, how are"
+
+ # Test that text is generator
+ self.assertGreater(len(model(start)), len(start))