[Inference Providers] Snippets: prefer the namespace/model:provider syntax for conversational (#1830)

# TL;DR

For conversational task, when using the Inference Clients libraries,
prefer the `namespace/model:provider` syntax over explicitly specifying
the provider in the parameters

Also don't specify `provider="auto"` because it's the default

Author: SBrandeis

packages/inference/src/snippets/getInferenceSnippets.ts

@@ -60,6 +60,8 @@ interface TemplateParams {
 	importBase64?: boolean; // specific to snippetImportRequests
 	importJson?: boolean; // specific to snippetImportRequests
 	endpointUrl?: string;
+	task?: InferenceTask;
+	directRequest?: boolean;
 }
 
 // Helpers to find + load templates
@@ -263,6 +265,8 @@ const snippetGenerator = (templateName: string, inputPreparationFn?: InputPrepar
 					: providerModelId ?? model.id,
 			billTo: opts?.billTo,
 			endpointUrl: opts?.endpointUrl,
+			task,
+			directRequest: !!opts?.directRequest,
 		};
 
 		/// Iterate over clients => check if a snippet exists => generate

packages/inference/src/snippets/templates/js/huggingface.js/conversational.jinja

@@ -6,8 +6,12 @@ const chatCompletion = await client.chatCompletion({
 {% if endpointUrl %}
     endpointUrl: "{{ endpointUrl }}",
 {% endif %}
+{% if directRequest %}
     provider: "{{ provider }}",
     model: "{{ model.id }}",
+{% else %}
+    model: "{{ providerModelId }}",
+{% endif %}
 {{ inputs.asTsString }}
 }{% if billTo %}, {
     billTo: "{{ billTo }}",

packages/inference/src/snippets/templates/js/huggingface.js/conversationalStream.jinja

@@ -8,8 +8,7 @@ const stream = client.chatCompletionStream({
 {% if endpointUrl %}
     endpointUrl: "{{ endpointUrl }}",
 {% endif %}
-    provider: "{{ provider }}",
-    model: "{{ model.id }}",
+    model: "{{ providerModelId }}",
 {{ inputs.asTsString }}
 }{% if billTo %}, {
     billTo: "{{ billTo }}",

packages/inference/src/snippets/templates/python/huggingface_hub/conversational.jinja

@@ -1,5 +1,9 @@
 completion = client.chat.completions.create(
+{% if directRequest %}
     model="{{ model.id }}",
+{% else %}
+    model="{{ providerModelId }}",
+{% endif %}
 {{ inputs.asPythonString }}
 )
 

packages/inference/src/snippets/templates/python/huggingface_hub/conversationalStream.jinja

@@ -1,5 +1,5 @@
 stream = client.chat.completions.create(
-    model="{{ model.id }}",
+    model="{{ providerModelId }}",
 {{ inputs.asPythonString }}
     stream=True,
 )

packages/inference/src/snippets/templates/python/huggingface_hub/importInferenceClient.jinja

@@ -4,7 +4,9 @@ client = InferenceClient(
 {% if endpointUrl %}
     base_url="{{ baseUrl }}",
 {% endif %}
+{% if task != "conversational" or directRequest %}
     provider="{{ provider }}",
+{% endif %}
     api_key="{{ accessToken }}",
 {% if billTo %}
     bill_to="{{ billTo }}",

packages/tasks-gen/snippets-fixtures/bill-to-param/js/huggingface.js/0.hf-inference.js

@@ -3,8 +3,7 @@ import { InferenceClient } from "@huggingface/inference";
 const client = new InferenceClient(process.env.HF_TOKEN);
 
 const chatCompletion = await client.chatCompletion({
-    provider: "hf-inference",
-    model: "meta-llama/Llama-3.1-8B-Instruct",
+    model: "meta-llama/Llama-3.1-8B-Instruct:hf-inference",
     messages: [
         {
             role: "user",

packages/tasks-gen/snippets-fixtures/bill-to-param/python/huggingface_hub/0.hf-inference.py

@@ -2,13 +2,12 @@
 from huggingface_hub import InferenceClient
 
 client = InferenceClient(
-    provider="hf-inference",
     api_key=os.environ["HF_TOKEN"],
     bill_to="huggingface",
 )
 
 completion = client.chat.completions.create(
-    model="meta-llama/Llama-3.1-8B-Instruct",
+    model="meta-llama/Llama-3.1-8B-Instruct:hf-inference",
     messages=[
         {
             "role": "user",

packages/tasks-gen/snippets-fixtures/conversational-llm-custom-endpoint/js/huggingface.js/0.hf-inference.js

@@ -4,7 +4,6 @@ const client = new InferenceClient(process.env.API_TOKEN);
 
 const chatCompletion = await client.chatCompletion({
     endpointUrl: "http://localhost:8080/v1",
-    provider: "hf-inference",
     model: "meta-llama/Llama-3.1-8B-Instruct",
     messages: [
         {

packages/tasks-gen/snippets-fixtures/conversational-llm-custom-endpoint/python/huggingface_hub/0.hf-inference.py

@@ -3,7 +3,6 @@
 
 client = InferenceClient(
     base_url="http://localhost:8080/v1",
-    provider="hf-inference",
     api_key=os.environ["API_TOKEN"],
 )