just in time install llama-cpp-python

ollama/engine.py

@@ -1,14 +1,13 @@
 import os
 import sys
 from os import path
-from pathlib import Path
 from contextlib import contextmanager
 from fuzzywuzzy import process
-from llama_cpp import Llama
 from ctransformers import AutoModelForCausalLM
 
 import ollama.prompt
 from ollama.model import MODELS_CACHE_PATH
+from ollama.platform import Llama
 
 
 @contextmanager

ollama/platform.py

@@ -0,0 +1,24 @@
+import os
+import sys
+import importlib
+from subprocess import call
+from contextlib import contextmanager
+
+
+@contextmanager
+def preserve_environ():
+    env = os.environ.copy()
+    yield
+    os.environ = env
+
+
+if not importlib.util.find_spec('llama_cpp'):
+    with preserve_environ():
+        os.environ['FORCE_CMAKE'] = '1'
+        if os.uname().sysname.lower() == 'darwin':
+            os.environ['CMAKE_ARGS'] = '-DLLAMA_METAL=on'
+
+        call([sys.executable, '-m', 'pip', 'install', 'llama-cpp-python==0.1.67'])
+
+
+from llama_cpp import Llama  # noqa: E402

poetry.lock

@@ -637,7 +637,7 @@ rapidfuzz = ">=2.3.0,<4.0.0"
 name = "llama-cpp-python"
 version = "0.1.67"
 description = "A Python wrapper for llama.cpp"
-optional = false
+optional = true
 python-versions = ">=3.7"
 files = [
     {file = "llama_cpp_python-0.1.67.tar.gz", hash = "sha256:33bdcd42b30df3c21d56ce094132e1cdc0da0f8a27109f8eaf698addad02fd20"},
@@ -1211,4 +1211,4 @@ termcolor = ">=2.2,<3.0"
 [metadata]
 lock-version = "2.0"
 python-versions = "^3.8"
-content-hash = "bd4b373e3903bd26b983163f3cc527a6f768f4280201fcbca4d4dc05dea66912"
+content-hash = "59f5dc1f3dbe73040cedac8d9484482f86f9ea8ae3793191de15014c0033debc"

pyproject.toml

@@ -16,10 +16,15 @@ requests = "^2.31.0"
 tqdm = "^4.65.0"
 validators = "^0.20.0"
 yaspin = "^2.3.0"
-llama-cpp-python = "^0.1.67"
 ctransformers = "^0.2.10"
 fuzzywuzzy = {extras = ["speedup"], version = "^0.18.0"}
 
+# required by llama-cpp-python
+llama-cpp-python = {version = "^0.1.67", optional = true}
+typing-extensions = "^4.6.3"
+numpy = "^1.24.4"
+diskcache = "^5.6.1"
+
 [build-system]
 requires = ["poetry-core"]
 build-backend = "poetry.core.masonry.api"

requirements.txt

@@ -385,8 +385,6 @@ levenshtein==0.21.1 ; python_version >= "3.8" and python_version < "4.0" \
     --hash=sha256:f5f7ce639bea0f5e95a1f71963624b85521a39928a2a1bb0e66f6180facf5969 \
     --hash=sha256:f9e3a5f4386c8f1811153f309a0ba3dc47d17e81a6dd29aa22d3e10212a2fd73 \
     --hash=sha256:ffa6762f8ef1e7dfba101babe43de6edc541cbe64d33d816314ac67cd76c3979
-llama-cpp-python==0.1.67 ; python_version >= "3.8" and python_version < "4.0" \
-    --hash=sha256:33bdcd42b30df3c21d56ce094132e1cdc0da0f8a27109f8eaf698addad02fd20
 markupsafe==2.1.3 ; python_version >= "3.8" and python_version < "4.0" \
     --hash=sha256:05fb21170423db021895e1ea1e1f3ab3adb85d1c2333cbc2310f2a26bc77272e \
     --hash=sha256:0a4e4a1aff6c7ac4cd55792abf96c915634c2b97e3cc1c7129578aa68ebd754e \