Mantisec/Ornith-1.5-9B-OBLITERATED-FP16

VISIONPricing:Input $0.431 / Cached $0.0862 / Output $1.12Concurrent Unit Cost:1Model Size:9BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Sep 1, 2026Architecture:Transformer Featherless Exclusive Cold

Mantisec/Ornith-1.5-9B-OBLITERATED-FP16 is a 9 billion parameter language model, converted by Mantisec using bfsquish v0.1.0 from OBLITERATUS/Ornith-1.5-9B-OBLITERATED. This model is specifically optimized for FP16 inference and fine-tuning on NVIDIA V100 (Volta) GPUs, which lack native BF16 Tensor Core support. It utilizes a range-checked conversion strategy to preserve numerical quality, making it suitable for environments requiring FP16 precision.

Loading preview...

Ornith-1.5-9B-OBLITERATED-FP16 Overview

This model is a 9 billion parameter language model, an FP16 conversion of OBLITERATUS/Ornith-1.5-9B-OBLITERATED performed by Mantisec using their bfsquish v0.1.0 tool. The primary purpose of this conversion is to enable efficient inference and fine-tuning on NVIDIA V100 (Volta) GPUs, which do not natively support BF16 Tensor Cores.

Key Capabilities

  • Optimized FP16 Performance: Specifically engineered for environments where FP16 precision is required or offers significant performance benefits, particularly on NVIDIA V100 GPUs.
  • Preserved Numerical Quality: The conversion employs a range_checked strategy, which directly upcasts to FP32 then performs a range-checked FP16 cast. This method ensures that values outside FP16's finite range are rejected, minimizing numerical degradation. Validation metrics show a token agreement rate of 99.81% and a minimum cosine similarity of 0.999951 against the source model.
  • Hardware Compatibility: Tailored for NVIDIA V100 (Volta, sm_70) hardware, allowing users with these GPUs to leverage the model effectively.

Good For

  • FP16 Inference: Ideal for deploying the 9B parameter Ornith-1.5 model in production environments that benefit from or require FP16 precision.
  • Fine-tuning on V100 GPUs: Developers and researchers with NVIDIA V100 GPUs can efficiently fine-tune this model without compatibility issues related to BF16.
  • Resource-constrained deployments: When memory or computational resources necessitate the use of FP16, this model provides a robust solution with validated numerical integrity.

Finetunes
1 models