update generate to yield object

ollama/cmd/cli.py

@@ -1,6 +1,5 @@
 import os
 import sys
-import json
 from pathlib import Path
 from argparse import ArgumentParser
 
@@ -67,7 +66,6 @@ def generate_oneshot(*args, **kwargs):
     print(flush=True)
 
     for output in engine.generate(*args, **kwargs):
-        output = json.loads(output)
         choices = output.get("choices", [])
         if len(choices) > 0:
             print(choices[0].get("text", ""), end="", flush=True)

ollama/cmd/server.py

@@ -1,5 +1,6 @@
-from aiohttp import web
+import json
 import aiohttp_cors
+from aiohttp import web
 
 from ollama import engine
 
@@ -89,7 +90,8 @@ async def generate(request):
     }
 
     for output in engine.generate(model, prompt, **kwargs):
-        await response.write(output.encode("utf-8"))
+        output = json.dumps(output).encode('utf-8')
+        await response.write(output)
         await response.write(b"\n")
 
     return response

ollama/engine.py

@@ -33,7 +33,7 @@ def generate(model, prompt, models_home=".", llms={}, *args, **kwargs):
         kwargs.update({"stream": True})
 
     for output in llm(prompt, *args, **kwargs):
-        yield json.dumps(output)
+        yield output
 
 
 def load(model, models_home=".", llms={}):