aasim-m/DAFT-Qwen3-0.6B-checkpoint-4000

TEXT GENERATIONPricing:Input $0.04 / Cached $0.008 / Output $0.08Concurrent Unit Cost:1Model Size:0.8BQuant:BF16Context Size:32kTool Calling:SupportedPublished:Sep 8, 2026License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Cold

The aasim-m/DAFT-Qwen3-0.6B-checkpoint-4000 is an intermediate fine-tuned checkpoint of the Qwen3-0.6B causal language model, featuring 0.8 billion parameters and a 32,768-token context length. Developed by aasim-m, this model is specifically optimized for translating NVIDIA GPU assembly functions into corresponding AMD GPU assembly. It serves as a specialized tool for developers working with GPU architecture translation tasks.

Loading preview...

Model Overview

This model, aasim-m/DAFT-Qwen3-0.6B-checkpoint-4000, is an intermediate full fine-tuning checkpoint of the Qwen3-0.6B base model, specifically designed for GPU assembly translation. It has 0.8 billion parameters and supports a substantial context length of 32,768 tokens.

Key Capabilities

  • GPU Assembly Translation: The primary function is to translate NVIDIA GPU assembly (cuda) into AMD GPU assembly (hip).
  • Specialized Fine-tuning: Fine-tuned on the ahmedheakl/daft-sm89-rdna-functions dataset, focusing exclusively on assembly translation pairs.
  • Intermediate Checkpoint: This is a snapshot taken at optimizer step 4,000 (0.39142 epochs) during an ongoing training run, indicating it's a work in progress.
  • High Context Window: Utilizes the Qwen3-0.6B's 32,768-token sequence limit for both input and output during translation.

Good For

  • GPU Architecture Migration: Ideal for developers and researchers needing to port or analyze NVIDIA GPU codebases for AMD architectures.
  • Specialized Code Translation: Useful for tasks requiring precise, low-level assembly code transformation.
  • Research and Development: Provides a foundation for further experimentation and fine-tuning in the domain of GPU assembly translation.

Limitations

  • Not General Purpose: This model is highly specialized and has not been evaluated as a general-purpose assistant.
  • Functional Correctness Not Assessed: Validation metrics provided are teacher-forced loss measurements, not functional correctness scores or generated-assembly accuracy. Downstream validation of generated assembly is required.

Finetunes
1 models