anthracite-org/magnum-v1-32b

TEXT GENERATIONConcurrency Cost:2Model Size:32.5BQuant:FP8Ctx Length:32kPublished:Jul 27, 2024License:tongyi-qianwenArchitecture:Transformer0.0K Cold

anthracite-org/magnum-v1-32b is a 32.5 billion parameter language model, fine-tuned on the Qwen1.5 32B architecture by the Anthracite team. It is specifically designed to replicate the prose quality of Claude 3 Sonnet and Opus models, excelling in prompt adherence and coherence. This model is optimized for high-quality prose generation and instruction following, making it suitable for applications requiring sophisticated text output.

Loading preview...

Model Overview

magnum-v1-32b is a 32.5 billion parameter language model developed by the Anthracite team, building upon the Qwen1.5 32B architecture. Its primary goal is to emulate the advanced prose quality found in Claude 3 Sonnet and Opus models.

Key Capabilities

  • High-Quality Prose Generation: Fine-tuned to produce text with a sophisticated and coherent style, similar to advanced commercial models.
  • Enhanced Prompt Adherence: Specifically designed to better follow instructions and maintain coherence within generated responses.
  • Instruction Following: The model has been instruction-tuned using ChatML formatting, making it responsive to detailed user prompts.

Training Details

The model underwent full-parameter fine-tuning for 2 epochs with a learning rate of 1e-05, utilizing 8x NVIDIA H100 Tensor Core GPUs. The training incorporated three new general-purpose instruction following datasets, including kalomaze/Opus_Instruct_25k, Nopm/Opus_WritingStruct, and a subset of Gryphe/Sonnet3.5-SlimOrcaDedupCleaned, to improve prompt adherence and writing structure.

Use Cases

This model is particularly well-suited for applications requiring high-fidelity text generation, creative writing, and complex instruction following where the quality and style of the output prose are critical.

Popular Sampler Settings

Top 3 parameter combinations used by Featherless users for this model. Click a tab to see each config.

temperature
top_p
top_k
frequency_penalty
presence_penalty
repetition_penalty
min_p