Fix/llms (#366)

* Fix LMStudio, cache documents with tg-load-sample-documents

* Fix Mistral

Makefile

@@ -67,6 +67,8 @@ some-containers:
 	    -t ${CONTAINER_BASE}/trustgraph-flow:${VERSION} .
 	${DOCKER} build -f containers/Containerfile.vertexai \
 	    -t ${CONTAINER_BASE}/trustgraph-vertexai:${VERSION} .
+	${DOCKER} build -f containers/Containerfile.bedrock \
+	    -t ${CONTAINER_BASE}/trustgraph-bedrock:${VERSION} .
 
 basic-containers: update-package-versions
 	${DOCKER} build -f containers/Containerfile.base \

trustgraph-cli/scripts/tg-invoke-llm

@@ -14,7 +14,7 @@ default_url = os.getenv("TRUSTGRAPH_URL", 'http://localhost:8088/')
 
 def query(url, flow_id, system, prompt):
 
-    api = Api(url).flow().id(flow_id).
+    api = Api(url).flow().id(flow_id)
 
     resp = api.text_completion(system=system, prompt=prompt)
 

trustgraph-cli/scripts/tg-load-sample-documents

@@ -11,7 +11,7 @@ import datetime
 import requests
 
 from trustgraph.api import Api
-from trustgraph.api.types import Uri, Literal, Triple
+from trustgraph.api.types import hash, Uri, Literal, Triple
 
 default_url = os.getenv("TRUSTGRAPH_URL", 'http://localhost:8088/')
 default_user = 'trustgraph'
@@ -29,6 +29,11 @@ session = requests.session()
 
 session.mount('file://', FileAdapter())
 
+try:
+    os.mkdir("doc-cache")
+except:
+    pass
+
 documents = [
 
     {
@@ -669,11 +674,19 @@ class Loader:
         try:
 
             print(doc["title"], ":")
-            print("  downloading...")
-            print("  done.")
 
-            resp = session.get(doc["url"])
-            content = resp.content
+            hid = hash(doc["url"])
+            cache_file = f"doc-cache/{hid}"
+
+            if os.path.isfile(cache_file):
+                print("  (use cache file)")
+                content = open(cache_file, "rb").read()
+            else:
+                print("  downloading...")
+                resp = session.get(doc["url"])
+                content = resp.content
+                open(cache_file, "wb").write(content)
+                print("  done.")
 
             print("  adding...")
 

trustgraph-flow/trustgraph/model/text_completion/cohere/llm.py

@@ -70,6 +70,8 @@ class Processor(LlmService):
                 model = self.model
             )
 
+            return resp
+
         # FIXME: Wrong exception, don't know what this LLM throws
         # for a rate limit
         except cohere.TooManyRequestsError:

trustgraph-flow/trustgraph/model/text_completion/lmstudio/llm.py

@@ -12,7 +12,6 @@ from .... base import LlmService, LlmResult
 
 default_ident = "text-completion"
 
-default_subscriber = module
 default_model = 'gemma3:9b'
 default_url = os.getenv("LMSTUDIO_URL", "http://localhost:1234/")
 default_temperature = 0.0

trustgraph-flow/trustgraph/model/text_completion/mistral/llm.py

@@ -86,6 +86,8 @@ class Processor(LlmService):
                 model = self.model
             )
 
+            return resp
+
         # FIXME: Wrong exception.  The MistralAI library has retry logic
         # so retry-able errors are retried transparently.  It means we
         # don't get rate limit events.