Feature/fastembed (#286)

* Copied in base files for fastembed

* Added fastembed implementation

* Added template support for fastembed

templates/components/embeddings-fastembed.jsonnet

@@ -0,0 +1,43 @@
+local base = import "base/base.jsonnet";
+local images = import "values/images.jsonnet";
+local url = import "values/url.jsonnet";
+local prompts = import "prompts/mixtral.jsonnet";
+
+{
+
+    "embeddings-model":: "sentence-transformers/all-MiniLM-L6-v2",
+
+    embeddings +: {
+    
+        create:: function(engine)
+
+            local container =
+                engine.container("embeddings")
+                    .with_image(images.trustgraph)
+                    .with_command([
+                        "embeddings-fastembed",
+                        "-p",
+                        url.pulsar,
+                        "-m",
+                        $["embeddings-model"],
+                    ])
+                    .with_limits("1.0", "400M")
+                    .with_reservations("0.5", "400M");
+
+            local containerSet = engine.containers(
+                "embeddings", [ container ]
+            );
+
+            local service =
+                engine.internalService(containerSet)
+                .with_port(8000, 8000, "metrics");
+
+            engine.resources([
+                containerSet,
+                service,
+            ])
+
+    },
+
+}
+

trustgraph-flow/scripts/embeddings-fastembed

@@ -0,0 +1,6 @@
+#!/usr/bin/env python3
+
+from trustgraph.embeddings.fastembed import run
+
+run()
+

trustgraph-flow/setup.py

@@ -41,6 +41,7 @@ setuptools.setup(
         "cohere",
         "cryptography",
         "falkordb",
+        "fastembed",
         "google-generativeai",
         "ibis",
         "jsonschema",
@@ -78,6 +79,7 @@ setuptools.setup(
         "scripts/document-embeddings",
         "scripts/document-rag",
         "scripts/embeddings-ollama",
+        "scripts/embeddings-fastembed",
         "scripts/ge-query-milvus",
         "scripts/ge-query-pinecone",
         "scripts/ge-query-qdrant",

trustgraph-flow/trustgraph/embeddings/fastembed/__init__.py

@@ -0,0 +1,3 @@
+
+from . processor import *
+

trustgraph-flow/trustgraph/embeddings/fastembed/__main__.py

@@ -0,0 +1,7 @@
+#!/usr/bin/env python3
+
+from . processor import run
+
+if __name__ == '__main__':
+    run()
+

trustgraph-flow/trustgraph/embeddings/fastembed/processor.py

@@ -0,0 +1,89 @@
+
+"""
+Embeddings service, applies an embeddings model selected from HuggingFace.
+Input is text, output is embeddings vector.
+"""
+
+from ... schema import EmbeddingsRequest, EmbeddingsResponse
+from ... schema import embeddings_request_queue, embeddings_response_queue
+from ... log_level import LogLevel
+from ... base import ConsumerProducer
+from fastembed import TextEmbedding
+import os
+
+module = ".".join(__name__.split(".")[1:-1])
+
+default_input_queue = embeddings_request_queue
+default_output_queue = embeddings_response_queue
+default_subscriber = module
+default_model="sentence-transformers/all-MiniLM-L6-v2"
+
+class Processor(ConsumerProducer):
+
+    def __init__(self, **params):
+
+        input_queue = params.get("input_queue", default_input_queue)
+        output_queue = params.get("output_queue", default_output_queue)
+        subscriber = params.get("subscriber", default_subscriber)
+
+        model = params.get("model", default_model)
+
+        super(Processor, self).__init__(
+            **params | {
+                "input_queue": input_queue,
+                "output_queue": output_queue,
+                "subscriber": subscriber,
+                "input_schema": EmbeddingsRequest,
+                "output_schema": EmbeddingsResponse,
+                "model": model,
+            }
+        )
+
+        self.embeddings = TextEmbedding(model_name = model)
+
+    def handle(self, msg):
+
+        v = msg.value()
+
+        # Sender-produced ID
+
+        id = msg.properties()["id"]
+
+        print(f"Handling input {id}...", flush=True)
+
+        text = v.text
+        vecs = self.embeddings.embed([text])
+
+        vecs = [
+            v.tolist()
+            for v in vecs
+        ]
+
+        print("Send response...", flush=True)
+        r = EmbeddingsResponse(
+            vectors=list(vecs),
+            error=None,
+        )
+
+        self.producer.send(r, properties={"id": id})
+
+        print("Done.", flush=True)
+
+    @staticmethod
+    def add_args(parser):
+
+        ConsumerProducer.add_args(
+            parser, default_input_queue, default_subscriber,
+            default_output_queue,
+        )
+
+        parser.add_argument(
+            '-m', '--model',
+            default=default_model,
+            help=f'Embeddings model (default: {default_model})'
+        )
+
+def run():
+
+    Processor.start(module, __doc__)
+

trustgraph-flow/trustgraph/embeddings/ollama/processor.py

@@ -1,6 +1,6 @@
 
 """
-Embeddings service, applies an embeddings model selected from HuggingFace.
+Embeddings service, applies an embeddings model hosted on a local Ollama.
 Input is text, output is embeddings vector.
 """