add prompt templates as j2 templates

ollama/cmd/cli.py

@@ -69,6 +69,7 @@ def generate_oneshot(*args, **kwargs):
         if len(choices) > 0:
             print(choices[0].get("text", ""), end="", flush=True)
 
+    # end with a new line
     print()
 
 

ollama/engine.py

@@ -3,9 +3,9 @@ import json
 import sys
 from contextlib import contextmanager
 from llama_cpp import Llama as LLM
-from template import template
 
 import ollama.model
+import ollama.prompt
 
 
 @contextmanager
@@ -21,7 +21,7 @@ def suppress_stderr():
 def generate(model, prompt, models_home=".", llms={}, *args, **kwargs):
     llm = load(model, models_home=models_home, llms=llms)
 
-    prompt = template(model, prompt)
+    prompt = ollama.prompt.template(model, prompt)
 
     if "max_tokens" not in kwargs:
         kwargs.update({"max_tokens": 16384})
@@ -43,12 +43,9 @@ def load(model, models_home=".", llms={}):
             name: path for name, path in ollama.model.models(models_home)
         }.get(model, None)
 
-        if model_path is None:
+        if not model_path:
             # try loading this as a path to a model, rather than a model name
-            if os.path.isfile(model):
-                model_path = model
-            else:
-                raise ValueError("Model not found")
+            model_path = model
 
         # suppress LLM's output
         with suppress_stderr():

ollama/prompt.py

@@ -0,0 +1,19 @@
+from os import path
+from difflib import SequenceMatcher
+from jinja2 import Environment, PackageLoader
+
+
+def template(model, prompt):
+    best_ratio = 0
+    best_template = ''
+
+    environment = Environment(loader=PackageLoader(__name__, 'templates'))
+    for template in environment.list_templates():
+        base, _ = path.splitext(template)
+        ratio = SequenceMatcher(None, model.lower(), base).ratio()
+        if ratio > best_ratio:
+            best_ratio = ratio
+            best_template = template
+
+    template = environment.get_template(best_template)
+    return template.render(prompt=prompt)

ollama/templates/alpaca.prompt

@@ -0,0 +1,8 @@
+Below is an instruction that describes a task. Write a response that appropriately completes the request.
+
+### Instruction:
+{{ prompt }}
+
+### Response:
+
+

ollama/templates/gpt4.prompt

@@ -0,0 +1,5 @@
+### Instruction:
+{{ prompt }}
+
+### Response:
+

ollama/templates/hermes.prompt

@@ -0,0 +1,5 @@
+### Instruction:
+{{ prompt }}
+
+### Response:
+

ollama/templates/oasst.prompt

@@ -0,0 +1 @@
+{{ prompt }}

ollama/templates/orca.prompt

@@ -0,0 +1,7 @@
+### System:
+You are an AI assistant that follows instruction extremely well. Help as much as you can.
+
+### User:
+{{ prompt }}
+
+### Response:

ollama/templates/qlora.prompt

@@ -0,0 +1,2 @@
+### Human: {{ prompt }}
+### Assistant:

ollama/templates/tulu.prompt

@@ -0,0 +1,4 @@
+
+{{ prompt }}
+
+

ollama/templates/vicuna.prompt

@@ -0,0 +1,4 @@
+A chat between a curious user and an artificial intelligence assistant. The assistant gives helpful, detailed, and polite answers to the user's questions.
+
+USER: {{ prompt }}
+ASSISTANT:

ollama/templates/wizardlm.prompt

@@ -0,0 +1,2 @@
+{{ prompt }}
+### Response: