add ollama support

guidance/models/__init__.py

@@ -3,6 +3,7 @@
 # local models
 from .transformers._transformers import Transformers, TransformersTokenizer
 from .llama_cpp import LlamaCpp
+from ._ollama import Ollama
 from ._mock import Mock, MockChat
 
 # grammarless models (we can't do constrained decoding for them)

guidance/models/_ollama.py

@@ -0,0 +1,49 @@
+import os
+import json
+
+from pathlib import Path
+
+from ._model import Model
+from .llama_cpp._llama_cpp import LlamaCppEngine
+
+base = Path(os.getenv('OLLAMA_MODELS', Path.home() / '.ollama' / 'models'))
+blobs = base / 'blobs'
+library =  base / 'manifests' / 'registry.ollama.ai' / 'library'
+
+
+class Ollama(Model):
+    def __init__(
+        self,
+        model: str,
+        echo=True,
+        compute_log_probs=False,
+        chat_template=None,
+        **llama_cpp_kwargs,
+    ):
+        """Wrapper for models pulled using Ollama.
+
+        Gets the local model path using the provided model name, and
+        then instantiates the `LlamaCppEngine` with it and other args.
+        """
+
+        manifest = library / Path(model.replace(':', '/') if ':' in model else model + '/latest')
+
+        if not manifest.exists():
+            raise ValueError(f"Model '{model}' not found in library.")
+
+        with open(manifest, 'r') as f:
+            for layer in json.load(f)['layers']:
+                if layer['mediaType'] == 'application/vnd.ollama.image.model':
+                    digest: str = layer['digest']
+                    break
+            else:
+                raise ValueError("Model layer not found in manifest.")
+
+        engine = LlamaCppEngine(
+            model=(blobs / digest.replace(':', '-')),
+            compute_log_probs=compute_log_probs,
+            chat_template=chat_template,
+            **llama_cpp_kwargs,
+        )
+
+        super().__init__(engine, echo=echo)