unconst/Affine-5czsc2fc98-r504-offline-dpo-hialpha-hirank-lobeta-midctx-extrasteps-merged

TEXT GENERATIONPricing:Input $0.4 / Cached $0.07 / Output $4Concurrent Unit Cost:3Model Size:35.1BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Aug 16, 2026Architecture:Transformer Featherless Exclusive Cold

The unconst/Affine-5czsc2fc98-r504-offline-dpo-hialpha-hirank-lobeta-midctx-extrasteps-merged model is a 35.1 billion parameter language model, merged from `kevin954/Affine-5dfqbbh8ev-sft`. With a context length of 32768 tokens, this model is a private checkpoint for internal development, not intended for general public use. Its primary characteristic is its origin as a salvaged LoRA-merged checkpoint, indicating a focus on specific, potentially experimental, fine-tuning applications.

Loading preview...

Overview

This model, unconst/Affine-5czsc2fc98-r504-offline-dpo-hialpha-hirank-lobeta-midctx-extrasteps-merged, is a substantial 35.1 billion parameter language model with an extended context length of 32768 tokens. It originates from a LoRA-merged checkpoint, specifically salvaged from kevin954/Affine-5dfqbbh8ev-sft.

Key Characteristics

  • Parameter Count: 35.1 billion parameters, indicating a large-scale model capable of complex language understanding and generation.
  • Context Length: Supports a significant context window of 32768 tokens, allowing it to process and generate longer, more coherent texts.
  • Origin: It is a LoRA-merged checkpoint, suggesting it has undergone efficient fine-tuning on a base model.
  • Development Status: Described as a "private TTL insurance" and "not a submission until Stage-5 gate clears," indicating it is an internal, developmental checkpoint rather than a publicly released or production-ready model.

Intended Use

  • Internal Development: Primarily intended for internal use and evaluation within its development lifecycle.
  • Experimental Fine-tuning: Given its LoRA-merged nature, it is likely used for exploring specific fine-tuning objectives or datasets.
  • Research & Salvage: Represents a salvaged checkpoint, potentially for further research, analysis, or as a base for subsequent iterations.