tweaks

Author: SBrandeis

packages/inference/src/lib/makeRequestOptions.ts

@@ -68,16 +68,16 @@ export async function makeRequestOptions(
 		}
 		switch (provider) {
 			case "replicate":
-				model = REPLICATE_MODEL_IDS[model];
+				model = REPLICATE_MODEL_IDS[model] ?? model;
 				break;
 			case "sambanova":
-				model = SAMBANOVA_MODEL_IDS[model];
+				model = SAMBANOVA_MODEL_IDS[model] ?? model;
 				break;
 			case "together":
 				model = TOGETHER_MODEL_IDS[model]?.id ?? model;
 				break;
 			case "fal-ai":
-				model = FAL_AI_MODEL_IDS[model];
+				model = FAL_AI_MODEL_IDS[model] ?? model;
 				break;
 			default:
 				break;

packages/inference/src/providers/replicate.ts

@@ -15,4 +15,7 @@ type ReplicateId = string;
 export const REPLICATE_MODEL_IDS: Record<ModelId, ReplicateId> = {
 	"black-forest-labs/FLUX.1-schnell": "black-forest-labs/flux-schnell",
 	"ByteDance/SDXL-Lightning": "bytedance/sdxl-lightning-4step",
+	"meta-llama/Meta-Llama-3-8B-Instruct": "meta/meta-llama-3-8b-instruct",
+	"meta-llama/Meta-Llama-3-70B-Instruct": "meta/meta-llama-3-70b-instruct",
+	"meta-llama/Meta-Llama-3.1-405B-Instruct": "meta/meta-llama-3.1-405b-instruct",
 };

packages/inference/src/tasks/custom/request.ts

@@ -26,9 +26,10 @@ export async function request<T>(
 	}
 
 	if (!response.ok) {
+		const contentType = response.headers.get("Content-Type");
 		if (
 			["application/json", "application/problem+json"].some(
-				(contentType) => response.headers.get("Content-Type")?.startsWith(contentType)
+				(ct) => contentType?.startsWith(ct)
 			)
 		) {
 			const output = await response.json();
@@ -41,7 +42,8 @@ export async function request<T>(
 				throw new Error(output);
 			}
 		}
-		throw new Error("An error occurred while fetching the blob");
+		const message = contentType?.startsWith("text/plain;") ? await response.text() : undefined;
+		throw new Error(message ?? "An error occurred while fetching the blob");
 	}
 
 	if (response.headers.get("Content-Type")?.startsWith("application/json")) {

packages/inference/src/tasks/nlp/chatCompletion.ts

@@ -11,6 +11,9 @@ export async function chatCompletion(
 	args: BaseArgs & ChatCompletionInput,
 	options?: Options
 ): Promise<ChatCompletionOutput> {
+	if (args.provider === "replicate") {
+		throw new Error("Replicate does not support the chat completion API");
+	}
 	const res = await request<ChatCompletionOutput>(args, {
 		...options,
 		taskHint: "text-generation",