response api

src-py/renatoxsr/scripts/prompt.py

@@ -32,6 +32,7 @@ write = write_rich
 
 def parse_args():
     p = argparse.ArgumentParser()
+    p.add_argument("--debug", action="store_true")
     p.add_argument("--model", 
             default="qwen3.5:4b",
             choices=[
@@ -46,19 +47,13 @@ def parse_args():
 
     mxg = p.add_mutually_exclusive_group()
     mxg.add_argument("--pull")
-    #mxg.add_argument("--chat", action=argparse.BooleanOptionalAction, default=True)
     mxg.add_argument("--response", action="store_true")
-    mxg.add_argument("--instruct", action="store_true")
+    mxg.add_argument("--endpoint")
-    mxg.add_argument("--endpoint",
-                   default="/api/chat",
-                   choices=[
-                       "/api/chat",
-                       "/v1/chat/completions",])
     mxg.add_argument("--api",
                    default="ollama",
                    choices=["ollama","openai",])
     g = p.add_argument_group("Model Options")
-    g.add_argument("--reasoning", action="store_true")
+    g.add_argument("--thinking", action=argparse.BooleanOptionalAction, default=True)
     g.add_argument("--stream", action="store_true")
     g.add_argument("--num_ctx", default=128000, type=int)
     g.add_argument("--top_k", default=40, type=int)
@@ -87,10 +82,10 @@ async def do_async_turn(conf):
     """Do user->assistant turn streaming responses"""
     if not conf.prompt:
         prompt = Prompt.ask(f"[{str(bb.model)}]Ask {conf.model}»[/] [{str(bb.prompt)}]")
-    if conf.response or conf.instruct:
+    if conf.response:
         system_and_user = {
-            "input": conf.prompt or prompt,
+            "input" if conf.api == "openai" else "prompt": conf.prompt or prompt,
-            "instructions": conf.system}
+            "instructions" if conf.api == "openai" else "system": conf.system}
     else:
         system_and_user = {"messages": [
             {"role":"system", "content": conf.system},
@@ -101,9 +96,14 @@ async def do_async_turn(conf):
             headers={
                 "Authorization": f"Bearer {conf.apikey}",},
                 timeout=float(conf.timeout))
-    async with client.stream('POST',"/api/chat", json=system_and_user | {
+    async with client.stream('POST',((
+            "/api/generate" if conf.response else "/api/chat")
+            if conf.api == "ollama" else ("/v1/responses"
+            if conf.response else "/v1/chat/completions")), 
+            json=system_and_user | {
             "model": conf.model,
-        "reasoning": {"enabled": conf.reasoning},
+            "reasoning": {"enabled": conf.thinking},
+            "thinking": conf.thinking,
             "stream": conf.stream,
             "options": {
                 "num_ctx": conf.num_ctx,
@@ -111,6 +111,13 @@ async def do_async_turn(conf):
                 "top_p": conf.top_p,
                 "temperature": conf.temperature,
             }}) as res:
+        res.raise_for_status()
+        if conf.debug:
+            write(res.request, end=' (')
+            write(", ".join([d for d in dir(res.request) if not d.startswith("_")]), end=")\n")
+            write(res.request.content)
+            write(res, end=' (')
+            write(", ".join([d for d in dir(res) if not d.startswith("_")]), end=')\n')
         msg = {"thinking": {"len": 1, 1: []},
                "content": {"len": 1, 1: []}}
         async for line in res.aiter_lines():
@@ -119,6 +126,8 @@ async def do_async_turn(conf):
                 # TODO: use spinner from rich
                 continue
             data = json.loads(line)
+            if conf.debug:
+                write(line)
             if data.get("done", False):
                 write(f"\n[Load: {s3f(data['load_duration'])}s | "
                       f"Analyze: {s3f(data['prompt_eval_duration'])}s | "
@@ -131,7 +140,7 @@ async def do_async_turn(conf):
                 msg['done'] = data
                 return msg
             #write(f"[i]{data}[/i]")
-            for i in ["thinking","content"]:
+            for i in ["thinking", "content"]:
                 text = data.get("message", {}).get(i, "")
                 msg[i][msg[i]['len']].append(text)
                 write(f"[{str(bb[i])}]{text}[/]", end='')     
@@ -141,13 +150,21 @@ async def do_async_turn(conf):
                     #write(Markdown("".join(msg[i][msg[i]['len']])))
                     msg[i]['len'] += 1
                     msg[i][msg[i]['len']] = []
+            write(data.get('choices',[{}])[0].get('message', ""), end='')
+            write(data.get('output', ""), end='')
+            write(f"[{str(bb.model)}]{data.get('response', '')}", end='')
+            write(f"[{str(bb.thinking)}]{data.get('thinking', '')}[/]", end='')
 
 def main():
     conf = parse_args()
     if conf.prompt == "-":
         conf.prompt = sys.stdin.read()
     if conf.prompt:
-        write(f"[{str(bb.model)}]Ask {conf.model}»[/] [{str(bb.prompt)}]{conf.prompt}[/]")
+        write(f"[{str(bb.model)}]Ask {conf.model}»[/] ", end='')
+        if conf.response:
+            write(f"[{str(bb.prompt)}]{conf.system}[/]")
+        else:
+            write(f"[{str(bb.prompt)}]{conf.prompt}[/]")
     try:
         while True:
             if conf.stream:
@@ -157,7 +174,8 @@ def main():
                 if "thinking" in data.get("message",{}):
                     write(f"[{str(bb.model)}]{conf.model}'s thinking:[/] [{str(bb.thinking)}] {Markdown(data['message']['thinking'])}[/]")
                 write(f"[{str(bb.model)}]{conf.model}'s response: [/][{str(bb.content)}]{Markdown(data['message']['content'])}[/]")
-                #response = response['choices'][0]['message']
+                write(data['choices'][0]['message'])
+                write(data['output'])
             if conf.prompt:
                 sys.exit(0)
             # TODO: add turn to message list and do another turn