fixes

bpm_ai_inference/llm/llama_cpp/_constants.py

@@ -1,5 +1,5 @@
 
-DEFAULT_MODEL = "QuantFactory/dolphin-2.9-llama3-8b-GGUF"
+DEFAULT_MODEL = "QuantFactory/Meta-Llama-3-8B-Instruct-GGUF"
 DEFAULT_QUANT_LARGE = "*Q8_0.gguf"
 DEFAULT_QUANT_BALANCED = "*Q4_K_M.gguf"
 DEFAULT_QUANT_SMALL = "*Q2_K.gguf"

bpm_ai_inference/llm/llama_cpp/llama_chat.py

@@ -65,6 +65,7 @@ def __init__(
                 verbose=False
             )
         else:
+            logger.info(f"Loading model file {filename} from {model}...")
             self.llm = Llama.from_pretrained(
                 repo_id=model,
                 filename=filename,

bpm_ai_inference/llm/llama_cpp/util.py

@@ -1,6 +1,7 @@
 import json
 import logging
 
+from bpm_ai_core.llm.common.blob import Blob
 from bpm_ai_core.llm.common.message import ChatMessage, AssistantMessage, ToolResultMessage
 
 logger = logging.getLogger(__name__)
@@ -18,6 +19,20 @@ async def message_to_llama_dict(message: ChatMessage) -> dict:
     elif isinstance(message, ToolResultMessage):
         tool_response_content = f"{message.content}"
         content = '<tool_response>\n{"name": "' + message.name + '", "content": ' + tool_response_content + '}\n</tool_response>'
+    elif isinstance(message.content, list):
+        content = ""
+        for e in message.content:
+            if isinstance(e, str):
+                content += e
+            elif isinstance(e, Blob) and (e.is_text()):
+                text = (await e.as_bytes()).decode("utf-8")
+                filename = (" name='" + e.path + "'") if e.path else ''
+                text = f"\n<file{filename}>\n{text}\n</file>\n"
+                content += text
+            else:
+                raise ValueError(
+                    "Elements in ChatMessage.content must be str or Blob (text)"
+                )
     else:
         content = message.content
 

pyproject.toml

@@ -1,6 +1,6 @@
 [tool.poetry]
 name = "bpm-ai-inference"
-version = "0.2.9"
+version = "0.3.1"
 description = "Inference and server for local AI implementations of bpm-ai-core abstractions."
 authors = ["Bennet Krause <[email protected]>"]
 repository = "https://github.com/holunda-io/bpm-ai-inference"
@@ -10,7 +10,7 @@ readme = "README.md"
 
 [tool.poetry.dependencies]
 python = ">=3.11,<3.12"
-bpm-ai-core = "^2.6.0"
+bpm-ai-core = "^2.6.1"
 langfuse = "^2.7.6"
 faster-whisper = "^0.10.0"
 lingua-language-detector = "^2.0.2"