Merge pull request #105 from fixie-ai/juberti/moar-funcs

Add Gemini and Groq function calling

llm_benchmark_suite.py

@@ -412,9 +412,11 @@ def _tools_models():
         _Llm("claude-3-5-sonnet-20240620"),
         _Llm("claude-3-sonnet-20240229"),
         _Llm("claude-3-haiku-20240307"),
+        _Llm(GEMINI_1_5_PRO),
+        _Llm(GEMINI_1_5_FLASH),
         _FireworksLlm("accounts/fireworks/models/firefunction-v2", "firefunction-v2"),
-        # _GroqLlm("llama3-groq-70b-8192-tool-use-preview"),
-        # _GroqLlm("llama3-groq-8b-8192-tool-use-preview"),
+        _GroqLlm("llama3-groq-70b-8192-tool-use-preview"),
+        _GroqLlm("llama3-groq-8b-8192-tool-use-preview"),
     ]
 
 

llm_request.py

@@ -250,10 +250,15 @@ async def openai_chunk_gen(response) -> TokenGenerator:
                 yield delta_content
             elif delta_tool:
                 function = delta_tool[0]["function"]
-                token = function.get("name") or function.get("arguments")
-                if token:
-                    yield token.strip()
-        usage = chunk.get("usage")
+                name = function.get("name", "").strip()
+                if name:
+                    tokens += 1
+                    yield name
+                args = function.get("arguments", "").strip()
+                if args:
+                    tokens += 1
+                    yield args
+        usage = chunk.get("usage") or chunk.get("x_groq", {}).get("usage")
         if usage:
             num_input_tokens = usage.get("prompt_tokens")
             num_output_tokens = usage.get("completion_tokens")
@@ -267,7 +272,7 @@ async def openai_chat(ctx: ApiContext, path: str = "/chat/completions") -> ApiRe
     kwargs = {"messages": make_openai_messages(ctx)}
     if ctx.tools:
         kwargs["tools"] = ctx.tools
-        kwargs["tool_choice"] = "auto"
+        kwargs["tool_choice"] = "required"
     if ctx.peft:
         kwargs["peft"] = ctx.peft
     data = make_openai_chat_body(ctx, **kwargs)
@@ -433,6 +438,14 @@ async def chunk_gen(response) -> TokenGenerator:
                     if "text" in part:
                         tokens += 1
                         yield part["text"]
+                    elif "functionCall" in part:
+                        call = part["functionCall"]
+                        if "name" in call:
+                            tokens += 1
+                            yield call["name"]
+                        if "args" in call:
+                            tokens += 1
+                            yield str(call["args"])
             usage = chunk.get("usageMetadata")
             if usage:
                 num_tokens = usage.get("candidatesTokenCount")
@@ -466,6 +479,10 @@ async def chunk_gen(response) -> TokenGenerator:
             if not ctx.files or ctx.files[0].is_image
         ],
     }
+    if ctx.tools:
+        data["tools"] = (
+            [{"function_declarations": [tool["function"] for tool in ctx.tools]}],
+        )
     return await post(ctx, url, headers, data, chunk_gen)
 
 
@@ -480,38 +497,6 @@ async def cohere_embed(ctx: ApiContext) -> ApiResult:
     return await post(ctx, url, headers, data)
 
 
-async def make_fixie_chunk_gen(response) -> TokenGenerator:
-    text = ""
-    async for line in response.content:
-        line = line.decode("utf-8").strip()
-        obj = json.loads(line)
-        curr_turn = obj["turns"][-1]
-        if (
-            curr_turn["role"] == "assistant"
-            and curr_turn["messages"]
-            and "content" in curr_turn["messages"][-1]
-        ):
-            if curr_turn["state"] == "done":
-                break
-            new_text = curr_turn["messages"][-1]["content"]
-            # Sometimes we get a spurious " " message
-            if new_text == " ":
-                continue
-            if new_text.startswith(text):
-                delta = new_text[len(text) :]
-                text = new_text
-                yield delta
-            else:
-                print(f"Warning: got unexpected text: '{new_text}' vs '{text}'")
-
-
-async def fixie_chat(ctx: ApiContext) -> ApiResult:
-    url = f"https://api.fixie.ai/api/v1/agents/{ctx.model}/conversations"
-    headers = make_headers(auth_token=get_api_key(ctx, "FIXIE_API_KEY"))
-    data = {"message": ctx.prompt, "runtimeParameters": {}}
-    return await post(ctx, url, headers, data, make_fixie_chunk_gen)
-
-
 async def fake_chat(ctx: ApiContext) -> ApiResult:
     class FakeResponse(aiohttp.ClientResponse):
         def __init__(self, status, reason):
@@ -613,7 +598,6 @@ def make_context(
             func = openai_chat
             if not args.base_url:
                 provider = "openai"
-        # case _ elif "/" in model return await fixie_chat(ctx)
         case _:
             raise ValueError(f"Unknown model: {model}")
     name = args.display_name or make_display_name(provider, model)