speaklar-llm-v3

Speaklar LLM v3

Speaklar is a conversational AI model designed for helpful Bengali, English, and Banglish interactions. It is created, developed, and trained by Munzur ul Mamun.

Speaklar is built for applications that need a clear, friendly conversational assistant with a consistent application identity and support for knowledge-grounded responses. The repository includes the merged full model, local GGUF variants, tokenizer and processor files. The model was trained on a curated conversational dataset combining grounded voicebot examples with identity-focused examples.

Identity

• Bangla name: স্পিকলার
• English name: Speaklar
• Creator: Munzur ul Mamun
• Developer: Munzur ul Mamun
• Trainer: Munzur ul Mamun

What Speaklar can be used for

Speaklar can be integrated into:

• Bengali customer-support chatbots
• Bangla and Banglish business assistants
• Website chat and product-help systems
• Voicebot and call-agent applications
• Social messaging assistants
• Knowledge-grounded question-answering tools
• Local and private inference workflows
• Prototyping, research, and conversational AI demonstrations

For product information and the Speaklar platform, visit speaklar.com.

Grounded responses

For factual applications, provide the relevant knowledge with the user question and instruct the assistant to answer only from that supplied context. When the requested information is absent, the application should instruct Speaklar to say that the information is unavailable instead of guessing.

The training data includes grounded Bengali voicebot conversations and identity-focused examples covering the name, creator, developer, trainer, introductions, informal wording, and misleading identity questions.

Available formats

The repository contains the full Transformers model and GGUF files for local inference. The GGUF files can be used with compatible local runtimes, including Ollama and llama.cpp-based applications.

Ollama versions are also available at ollama.com/munzuruleee/speaklar-llm-v3.

Demo

Try Speaklar in the hosted demo:

Speaklar LLM v3 Demo

The demo includes a system-prompt field and a knowledge-context field for testing controlled, grounded responses.

Example questions

তোমার নাম কী?
Who created you?
তোমাকে কে train করেছে?
নিজের পরিচয় দাও।

Notes

Use an explicit system prompt and relevant knowledge context in production applications. Validate generated answers for your use case, especially when handling business, account, order, or other consequential information.