Merge pull request #122 from fixie-ai/juberti/mistral

Add Mistral Large 2 and Nemo

Author: juberti

llm_benchmark_suite.py

@@ -40,7 +40,6 @@
 LLAMA_3_8B_CHAT_FP4 = "llama-3-8b-chat-fp4"
 MIXTRAL_8X7B_INSTRUCT = "mixtral-8x7b-instruct"
 MIXTRAL_8X7B_INSTRUCT_FP8 = "mixtral-8x7b-instruct-fp8"
-PHI_2 = "phi-2"
 
 
 parser = argparse.ArgumentParser()
@@ -183,6 +182,18 @@ def __init__(self, model: str, display_model: Optional[str] = None):
         )
 
 
+class _MistralLlm(_Llm):
+    """See https://docs.mistral.ai/getting-started/models"""
+
+    def __init__(self, model: str, display_model: Optional[str] = None):
+        super().__init__(
+            model,
+            "mistral.ai/" + (display_model or model),
+            api_key=os.getenv("MISTRAL_API_KEY"),
+            base_url="https://api.mistral.ai/v1",
+        )
+
+
 class _NvidiaLlm(_Llm):
     """See https://build.nvidia.com/explore/discover"""
 
@@ -330,6 +341,9 @@ def _text_models():
         _Llm("gemini-pro"),
         _Llm(GEMINI_1_5_PRO),
         _Llm(GEMINI_1_5_FLASH),
+        # Mistral
+        _MistralLlm("mistral-large-latest", "mistral-large"),
+        _MistralLlm("open-mistral-nemo", "mistral-nemo"),
         # Mistral 8x7b
         _DatabricksLlm("databricks-mixtral-8x7b-instruct", MIXTRAL_8X7B_INSTRUCT),
         _DeepInfraLlm("mistralai/Mixtral-8x7B-Instruct-v0.1", MIXTRAL_8X7B_INSTRUCT),
@@ -484,6 +498,7 @@ def _image_models():
         _FireworksLlm(
             "accounts/fireworks/models/phi-3-vision-128k-instruct", "phi-3-vision"
         ),
+        _MistralLlm("pixtral-latest", "pixtral"),
     ]
 
 

llm_request.py

@@ -146,7 +146,9 @@ async def run(self, on_token: Optional[Callable[["ApiContext", str], None]] = No
         if not self.metrics.error:
             token_time = end_time - first_token_time
             self.metrics.total_time = end_time - start_time
-            self.metrics.tps = min((self.metrics.output_tokens - 1) / token_time, MAX_TPS)
+            self.metrics.tps = min(
+                (self.metrics.output_tokens - 1) / token_time, MAX_TPS
+            )
             if self.metrics.tps == MAX_TPS:
                 self.metrics.tps = 0.0
         else:
@@ -293,7 +295,7 @@ async def openai_chat(ctx: ApiContext, path: str = "/chat/completions") -> ApiRe
     # Some providers require opt-in for stream stats, but some providers don't like this opt-in.
     # Regardless of opt-in, Azure and ovh.net don't return stream stats at the moment.
     # See https://github.com/Azure/azure-rest-api-specs/issues/25062
-    if not any(p in ctx.name for p in ["azure", "databricks", "fireworks"]):
+    if not any(p in ctx.name for p in ["azure", "databricks", "fireworks", "mistral"]):
         kwargs["stream_options"] = {"include_usage": True}
     data = make_openai_chat_body(ctx, **kwargs)
     return await post(ctx, url, headers, data, openai_chunk_gen)