refactor: use async-openai CompletionRequest (NVIDIA#310)

Author: paulhendricks

lib/llm/src/http/service/openai.rs

@@ -140,17 +140,19 @@ async fn completions(
     let request_id = uuid::Uuid::new_v4().to_string();
 
     // todo - decide on default
-    let streaming = request.stream.unwrap_or(false);
+    let streaming = request.inner.stream.unwrap_or(false);
 
     // update the request to always stream
-    let request = CompletionRequest {
+    let inner = async_openai::types::CreateCompletionRequest {
         stream: Some(true),
-        ..request
+        ..request.inner
     };
 
+    let request = CompletionRequest { inner, nvext: None };
+
     // todo - make the protocols be optional for model name
     // todo - when optional, if none, apply a default
-    let model = &request.model;
+    let model = &request.inner.model;
 
     // todo - error handling should be more robust
     let engine = state

lib/llm/src/preprocessor/prompt/template/oai.rs

@@ -60,7 +60,7 @@ impl OAIChatLikeRequest for CompletionRequest {
         let message = async_openai::types::ChatCompletionRequestMessage::User(
             async_openai::types::ChatCompletionRequestUserMessage {
                 content: async_openai::types::ChatCompletionRequestUserMessageContent::Text(
-                    self.prompt.clone(),
+                    crate::protocols::openai::completions::prompt_to_string(&self.inner.prompt),
                 ),
                 name: None,
             },

lib/llm/src/protocols/openai.rs

@@ -22,11 +22,9 @@ pub mod nvext;
 use anyhow::Result;
 use serde::{Deserialize, Serialize};
 use std::{
-    collections::HashMap,
     fmt::Display,
     ops::{Add, Div, Mul, Sub},
 };
-use validator::ValidationError;
 
 use super::{
     common::{self, SamplingOptionsProvider, StopConditionsProvider},
@@ -263,17 +261,6 @@ pub struct GenericCompletionResponse<C>
     // TODO() - add NvResponseExtention
 }
 
-fn validate_logit_bias(logit_bias: &HashMap<String, i32>) -> Result<(), ValidationError> {
-    for key in logit_bias.keys() {
-        if key.parse::<i32>().is_err() {
-            return Err(
-                ValidationError::new("logit_bias").with_message("Keys must be integers".into())
-            );
-        }
-    }
-    Ok(())
-}
-
 // todo - move to common location
 fn validate_range<T>(value: Option<T>, range: &(T, T)) -> Result<Option<T>>
 where