srirag/alexmt-qwen3-0.6b-sft

TEXT GENERATIONPricing:Input $0.04 / Cached $0.008 / Output $0.08Concurrent Unit Cost:1Model Size:0.8BQuant:BF16Context Size:32kTool Calling:SupportedPublished:Sep 28, 2026License:cc-by-nc-4.0Architecture:Transformer Open Weights Featherless Exclusive Cold

The srirag/alexmt-qwen3-0.6b-sft model is a 0.8 billion parameter Qwen3-based language model fine-tuned for context-aware English to dialectal Arabic and dialectal Arabic to English translation. It specializes in conversational turns across nine Arabic varieties, leveraging the Alexandria dataset for supervised fine-tuning. This model is designed to provide culturally inclusive and linguistically diverse machine translation, particularly for dialectal Arabic in conversational contexts.

Loading preview...

Model Overview

The srirag/alexmt-qwen3-0.6b-sft is a 0.8 billion parameter model based on the Qwen3 architecture, specifically fine-tuned for context-aware English ↔ dialectal Arabic translation in conversational settings. Developed by srirag, this model addresses the challenge of translating between English and various dialectal Arabic forms, considering conversational context for improved accuracy.

Key Capabilities

  • Bidirectional Translation: Supports translation from English to dialectal Arabic and from dialectal Arabic to English.
  • Context-Aware: Utilizes up to three prior turns in a conversation as context to inform translations, enhancing relevance and coherence.
  • Dialectal Coverage: Trained on nine distinct Arabic varieties from the Alexandria dataset, including Egyptian (EG), Jordanian (JO), Lebanese (LB), Moroccan (MR), Omani (OM), Palestinian (PS), Saudi (SA), Syrian (SY), and Yemeni (YE).
  • Conversational Focus: Optimized for translating individual turns within a dialogue, considering domain, participants, and speaker/addressee gender direction.

Performance and Usage

This model was fine-tuned using supervised learning (full fine-tune, bf16, 2 epochs) on the Alexandria train split. It achieves a macro-average spBLEU of 9.17 for English to dialectal Arabic and 24.49 for dialectal Arabic to English on the Alexandria public test split. Users should apply the tokenizer's chat template with add_generation_prompt=True and provide input in Alexandria's JSON prompt format, expecting a {"translation": "..."} output. The model's license is CC BY-NC 4.0, inherited from its training data.

Finetunes
1 models