fix log probs

llama_stack/apis/inference/client.py

@@ -6,7 +6,6 @@
 
 import asyncio
 import json
-import sys
 from typing import Any, AsyncGenerator, List, Optional
 
 import fire
@@ -101,7 +100,9 @@ class InferenceClient(Inference):
                             print(f"Error with parsing or validation: {e}")
 
 
-async def run_main(host: str, port: int, stream: bool, model: Optional[str]):
+async def run_main(
+    host: str, port: int, stream: bool, model: Optional[str], logprobs: bool
+):
     client = InferenceClient(f"http://{host}:{port}")
 
     if not model:
@@ -115,7 +116,13 @@ async def run_main(host: str, port: int, stream: bool, model: Optional[str]):
         model=model,
         messages=[message],
         stream=stream,
+        logprobs=logprobs,
     )
+
+    if logprobs:
+        async for chunk in iterator:
+            cprint(f"Response: {chunk}", "red")
+    else:
         async for log in EventLogger().log(iterator):
             log.print()
 
@@ -149,13 +156,14 @@ def main(
     port: int,
     stream: bool = True,
     mm: bool = False,
+    logprobs: bool = False,
     file: Optional[str] = None,
     model: Optional[str] = None,
 ):
     if mm:
         asyncio.run(run_mm_main(host, port, stream, file, model))
     else:
-        asyncio.run(run_main(host, port, stream, model))
+        asyncio.run(run_main(host, port, stream, model, logprobs))
 
 
 if __name__ == "__main__":

llama_stack/apis/inference/inference.py

@@ -69,7 +69,7 @@ class ChatCompletionResponseEvent(BaseModel):
 
     event_type: ChatCompletionResponseEventType
     delta: Union[str, ToolCallDelta]
-    logprobs: Optional[List[TokenLogProbs]] = None
+    logprobs: Optional[List[float]] = None
     stop_reason: Optional[StopReason] = None
 
 
@@ -80,7 +80,7 @@ class CompletionRequest(BaseModel):
     sampling_params: Optional[SamplingParams] = SamplingParams()
 
     stream: Optional[bool] = False
-    logprobs: Optional[LogProbConfig] = None
+    logprobs: Optional[bool] = False
 
 
 @json_schema_type
@@ -88,7 +88,7 @@ class CompletionResponse(BaseModel):
     """Completion response."""
 
     completion_message: CompletionMessage
-    logprobs: Optional[List[TokenLogProbs]] = None
+    logprobs: Optional[List[float]] = None
 
 
 @json_schema_type
@@ -97,7 +97,7 @@ class CompletionResponseStreamChunk(BaseModel):
 
     delta: str
     stop_reason: Optional[StopReason] = None
-    logprobs: Optional[List[TokenLogProbs]] = None
+    logprobs: Optional[List[float]] = None
 
 
 @json_schema_type
@@ -105,7 +105,7 @@ class BatchCompletionRequest(BaseModel):
     model: str
     content_batch: List[InterleavedTextMedia]
     sampling_params: Optional[SamplingParams] = SamplingParams()
-    logprobs: Optional[LogProbConfig] = None
+    logprobs: Optional[bool] = False
 
 
 @json_schema_type
@@ -129,7 +129,7 @@ class ChatCompletionRequest(BaseModel):
     )
 
     stream: Optional[bool] = False
-    logprobs: Optional[LogProbConfig] = None
+    logprobs: Optional[bool] = False
 
 
 @json_schema_type
@@ -144,7 +144,7 @@ class ChatCompletionResponse(BaseModel):
     """Chat completion response."""
 
     completion_message: CompletionMessage
-    logprobs: Optional[List[TokenLogProbs]] = None
+    logprobs: Optional[List[float]] = None
 
 
 @json_schema_type
@@ -159,7 +159,7 @@ class BatchChatCompletionRequest(BaseModel):
     tool_prompt_format: Optional[ToolPromptFormat] = Field(
         default=ToolPromptFormat.json
     )
-    logprobs: Optional[LogProbConfig] = None
+    logprobs: Optional[bool] = False
 
 
 @json_schema_type
@@ -180,7 +180,7 @@ class Inference(Protocol):
         content: InterleavedTextMedia,
         sampling_params: Optional[SamplingParams] = SamplingParams(),
         stream: Optional[bool] = False,
-        logprobs: Optional[LogProbConfig] = None,
+        logprobs: Optional[bool] = None,
     ) -> Union[CompletionResponse, CompletionResponseStreamChunk]: ...
 
     @webmethod(route="/inference/chat_completion")
@@ -194,7 +194,7 @@ class Inference(Protocol):
         tool_choice: Optional[ToolChoice] = ToolChoice.auto,
         tool_prompt_format: Optional[ToolPromptFormat] = ToolPromptFormat.json,
         stream: Optional[bool] = False,
-        logprobs: Optional[LogProbConfig] = None,
+        logprobs: Optional[bool] = None,
     ) -> Union[ChatCompletionResponse, ChatCompletionResponseStreamChunk]: ...
 
     @webmethod(route="/inference/embeddings")

llama_stack/distribution/routers/routers.py

@@ -101,7 +101,7 @@ class InferenceRouter(Inference):
         tool_choice: Optional[ToolChoice] = ToolChoice.auto,
         tool_prompt_format: Optional[ToolPromptFormat] = ToolPromptFormat.json,
         stream: Optional[bool] = False,
-        logprobs: Optional[LogProbConfig] = None,
+        logprobs: Optional[bool] = False,
     ) -> AsyncGenerator:
         params = dict(
             model=model,
@@ -125,7 +125,7 @@ class InferenceRouter(Inference):
         content: InterleavedTextMedia,
         sampling_params: Optional[SamplingParams] = SamplingParams(),
         stream: Optional[bool] = False,
-        logprobs: Optional[LogProbConfig] = None,
+        logprobs: Optional[bool] = False,
     ) -> Union[CompletionResponse, CompletionResponseStreamChunk]:
         return await self.routing_table.get_provider_impl(model).completion(
             model=model,

llama_stack/providers/impls/meta_reference/inference/generation.py

@@ -297,7 +297,7 @@ class Llama:
                 token=next_token[0].item(),
                 text=self.tokenizer.decode(next_token.tolist()),
                 logprobs=(
-                    token_logprobs[:, prev_pos + 1 : cur_pos + 1][0].tolist()
+                    token_logprobs[:, cur_pos : cur_pos + 1][0].tolist()
                     if logprobs
                     else None
                 ),

llama_stack/providers/impls/meta_reference/inference/inference.py

@@ -58,7 +58,7 @@ class MetaReferenceInferenceImpl(Inference, RoutableProvider):
         tool_choice: Optional[ToolChoice] = ToolChoice.auto,
         tool_prompt_format: Optional[ToolPromptFormat] = ToolPromptFormat.json,
         stream: Optional[bool] = False,
-        logprobs: Optional[LogProbConfig] = None,
+        logprobs: Optional[bool] = None,
     ) -> AsyncIterator[
         Union[ChatCompletionResponseStreamChunk, ChatCompletionResponse]
     ]:
@@ -132,7 +132,10 @@ class MetaReferenceInferenceImpl(Inference, RoutableProvider):
 
                 if not request.stream:
                     if request.logprobs:
-                        logprobs.append(token_result.logprob)
+                        assert (
+                            len(token_result.logprobs) == 1
+                        ), "Expected logprob to contain 1 result for the current token"
+                        logprobs.append(token_result.logprobs[0])
 
                     continue