Task 'text-generation' not supported for provider PROVIDER. Available tasks: conversational

[andruhon]

Checked other resources

• I added a very descriptive title to this question.
• I searched the LangChain documentation with the integrated search.
• I used the GitHub search to find a similar question and didn't find it.

Commit to Help

• I commit to help with one of those options 👆

Example Code

import { HuggingFaceInference } from '@langchain/community/llms/hf';
import { BufferMemory } from 'langchain/memory';
import { ConversationChain } from 'langchain/chains';
import { PromptTemplate } from '@langchain/core/prompts';

// // This somewhat works ( see https://js.langchain.com/docs/integrations/llms/huggingface_inference/ )
const model1 = new HuggingFaceInference({
  model: 'mistralai/Mistral-Nemo-Base-2407',
});

const res1 = await model1.invoke("Hi!");
console.log({ res1 });

// This one gives "InferenceClientInputError: Task 'text-generation' not supported for provider 'fireworks-ai'. Available tasks: conversational"
const model2 = new HuggingFaceInference({
  model: 'mistralai/Mixtral-8x22B-Instruct-v0.1',
});
const res2 = await model2.invoke("What is the capital of Norway?");
console.log({ res2 });

// For me a logical assumption was to use ConversationChain, but no luck

// This one gives the same "Task 'text-generation' not supported for provider 'fireworks-ai'"
const model3 = new HuggingFaceInference({
  model: 'mistralai/Mixtral-8x22B-Instruct-v0.1',
});
const prompt = PromptTemplate.fromTemplate("You are a helpful agent.")
const memory = new BufferMemory({ memoryKey: 'chat_history' });
const chain = new ConversationChain({
  llm: model3,
  prompt,
  memory
});
const res3 = await chain.call({ input: 'What is the capital of Norway?' });
console.log({ res3 });

Description

I am trying to figure out how to use huggingface with LangChain.js, but I have some difficulties with most models responding that they only have conversational type of tasks availab.

I had a look at dist, it ends up in InferenceClient.js of huggingface, which does not seem to have conversational in the list of tasks.

Mistrai here is just as an example, tried deepseek ai with the same outcome.

I searched for similar questions, but I didn't find the answer there, and also they are 2 years old already:
#3256
#2070

System Info

"@langchain/core": "^0.3.58",
"@langchain/langgraph": "^0.3.1",
"@huggingface/inference": "^4.1.0",
"@langchain/community": "^0.3.47",

[dosubot[bot]]

Hi @andruhon! I'm Dosu and I’m helping the langchainjs team.

The error happens because the HuggingFaceInference integration in LangChain.js always tries to use the "text-generation" task, but some providers (like "fireworks-ai") only support the "conversational" task for certain models (such as mistralai/Mixtral-8x22B-Instruct-v0.1). There’s currently no user-facing option in LangChain.js to override or force the task type for LLMs—the task is determined internally based on the model and provider metadata from the Hugging Face Inference API. This is why models like Mistral-Nemo-Base-2407 work, but others do not, depending on what the provider exposes for that model code reference.

Recent updates allow specifying a provider for embeddings, but this is not yet exposed for LLMs in the HuggingFaceInference wrapper PR reference. There’s also no documented workaround for forcing the "conversational" task or switching providers for LLMs.

If you need to use a model that only supports "conversational" with a specific provider, you’ll need to wait for LangChain.js to expose more advanced configuration, or try using a different model/provider combination that supports "text-generation" for your use case.

To reply, just mention @dosu.

---

^{How did I do? Good | Irrelevant | Incorrect | Verbose | Hallucination | Report 🐛 | Other}   

[andruhon]

Thank you Dosubot.

Just as a note: calling this with @huggingface/inference directly works:

import { InferenceClient } from '@huggingface/inference';

const client = new InferenceClient(process.env.HUGGINGFACEHUB_API_KEY);
const out = await client.chatCompletion({
  model: 'deepseek-ai/DeepSeek-R1',
  messages: [{ role: 'user', content: 'What does JS stand for?' }],
  max_tokens: 512,
});
console.log(out.choices[0].message);