Merge remote-tracking branch 'upstream/main'

Author: nbarr07

README.md

@@ -97,7 +97,7 @@ You can run our packages with vanilla JS, without any bundler, by using a CDN or
 
 ```html
 <script type="module">
-    import { InferenceClient } from 'https://cdn.jsdelivr.net/npm/@huggingface/[email protected].0/+esm';
+    import { InferenceClient } from 'https://cdn.jsdelivr.net/npm/@huggingface/[email protected].1/+esm';
     import { createRepo, commit, deleteRepo, listFiles } from "https://cdn.jsdelivr.net/npm/@huggingface/[email protected]/+esm";
 </script>
 ```

packages/inference/package.json

@@ -1,6 +1,6 @@
 {
 	"name": "@huggingface/inference",
-	"version": "3.7.0",
+	"version": "3.7.1",
 	"packageManager": "[email protected]",
 	"license": "MIT",
 	"author": "Hugging Face and Tim Mikeladze <[email protected]>",

packages/inference/src/snippets/getInferenceSnippets.ts

@@ -254,7 +254,7 @@ const prepareConversationalInput = (
 	return {
 		messages: opts?.messages ?? getModelInputSnippet(model),
 		...(opts?.temperature ? { temperature: opts?.temperature } : undefined),
-		max_tokens: opts?.max_tokens ?? 500,
+		max_tokens: opts?.max_tokens ?? 512,
 		...(opts?.top_p ? { top_p: opts?.top_p } : undefined),
 	};
 };

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

@@ -5,18 +5,12 @@ const client = new OpenAI({
 	apiKey: "{{ accessToken }}",
 });
 
-let out = "";
-
 const stream = await client.chat.completions.create({
-    provider: "{{ provider }}",
-    model: "{{ model.id }}",
+    model: "{{ providerModelId }}",
 {{ inputs.asTsString }}
+    stream: true,
 });
 
 for await (const chunk of stream) {
-	if (chunk.choices && chunk.choices.length > 0) {
-		const newContent = chunk.choices[0].delta.content;
-		out += newContent;
-		console.log(newContent);
-	}  
+    process.stdout.write(chunk.choices[0]?.delta?.content || "");
 }

packages/ollama-utils/package.json

@@ -1,7 +1,7 @@
 {
 	"name": "@huggingface/ollama-utils",
 	"packageManager": "[email protected]",
-	"version": "0.0.7",
+	"version": "0.0.8",
 	"description": "Various utilities for maintaining Ollama compatibility with models on Hugging Face hub",
 	"repository": "https://github.com/huggingface/huggingface.js.git",
 	"publishConfig": {

packages/ollama-utils/src/chat-template-automap.ts

@@ -81,6 +81,16 @@ const CUSTOM_TEMPLATE_MAPPING: ((ggufTmpl: string) => OllamaCustomMappedTemplate
 					stop: "<|END_OF_TURN_TOKEN|>",
 			  }
 			: undefined,
+	(ggufTmpl: string) =>
+		ggufTmpl.match(/Mistral Small 3/) && ggufTmpl.match(/2023-10-01/)
+			? {
+					// https://ollama.com/library/mistral-small
+					// (template is edited at some point, so we need manual map to make sure it works, ref commit: https://huggingface.co/mistralai/Mistral-Small-3.1-24B-Instruct-2503/commit/4b8dd8aae705887db5295fcbff4aedbb92d682eb)
+					ollamaTmpl:
+						'{{- range $index, $_ := .Messages }}\n{{- if eq .Role "system" }}[SYSTEM_PROMPT]{{ .Content }}[/SYSTEM_PROMPT]\n{{- else if eq .Role "user" }}\n{{- if and (le (len (slice $.Messages $index)) 2) $.Tools }}[AVAILABLE_TOOLS]{{ $.Tools }}[/AVAILABLE_TOOLS]\n{{- end }}[INST]{{ .Content }}[/INST]\n{{- else if eq .Role "assistant" }}\n{{- if .Content }}{{ .Content }}\n{{- if not (eq (len (slice $.Messages $index)) 1) }}</s>\n{{- end }}\n{{- else if .ToolCalls }}[TOOL_CALLS][\n{{- range .ToolCalls }}{"name": "{{ .Function.Name }}", "arguments": {{ .Function.Arguments }}}\n{{- end }}]</s>\n{{- end }}\n{{- else if eq .Role "tool" }}[TOOL_RESULTS]{"content": {{ .Content }}}[/TOOL_RESULTS]\n{{- end }}\n{{- end }}',
+					stop: "[INST]",
+			  }
+			: undefined,
 ];
 
 export function convertGGUFTemplateToOllama(

packages/ollama-utils/src/chat-template.ts

@@ -11,7 +11,7 @@ const chatCompletion = await client.chatCompletion({
             content: "What is the capital of France?",
         },
     ],
-    max_tokens: 500,
+    max_tokens: 512,
 });
 
 console.log(chatCompletion.choices[0].message);

packages/tasks-gen/snippets-fixtures/conversational-llm-non-stream/js/huggingface.js/0.hf-inference.js

@@ -11,7 +11,7 @@ const chatCompletion = await client.chatCompletion({
             content: "What is the capital of France?",
         },
     ],
-    max_tokens: 500,
+    max_tokens: 512,
 });
 
 console.log(chatCompletion.choices[0].message);

packages/tasks-gen/snippets-fixtures/conversational-llm-non-stream/js/huggingface.js/0.together.js

@@ -13,7 +13,7 @@ const chatCompletion = await client.chat.completions.create({
             content: "What is the capital of France?",
         },
     ],
-    max_tokens: 500,
+    max_tokens: 512,
 });
 
 console.log(chatCompletion.choices[0].message);