This project distills the inference ability of GPT into the Llama-3.1-8B model through an innovative two-stage training process. First, knowledge distillation and format alignment are performed through supervised fine-tuning, and then reinforcement learning is used to encourage the model to independently explore and optimize inference strategies, focusing on breakthroughs in the field of mathematical reasoning.

Natural Language ProcessingTransformersMultiple Languages

This model is based on GPT-OSS-20B and fine-tuned using the Unsloth reinforcement learning framework. The aim is to optimize inference efficiency and reduce vulnerabilities that occur during reinforcement learning from human feedback (RLHF) training. The fine-tuning process focuses on the robustness and efficiency of alignment, ensuring that the model maintains inference depth without incurring excessive computational overhead.

Natural Language ProcessingTransformersEnglish

GPT-OSS-20b-GGUF is a quantized version based on the OpenAI GPT-OSS-20b model, using the Apache 2.0 license agreement with lenient usage terms. This model supports text generation tasks, has a complete thought chain inference ability, and is natively quantized with MXFP4, enabling efficient operation under different memory conditions.

Natural Language ProcessingGguf

A quantized version of the OpenAI GPT-OSS-120B model, quantized using the llama.cpp tool and the imatrix dataset. It offers multiple quantization levels to meet different hardware requirements and supports running in the LM Studio and llama.cpp environments.

Natural Language ProcessingGguf

A large language model optimized based on OpenAI's gpt-oss-120b model. It uses the NEO dataset to improve performance, supports multi-scenario applications, and can run on CPU, GPU, or in partial offloading mode. It has a 128k context length and a mixture of experts architecture with up to 128 experts.

Natural Language ProcessingGgufMultiple Languages

Tiny GptOssForCausalLM is a minimized causal language model designed specifically for unit tests of the TRL (Transformer Reinforcement Learning) library. This model is extremely small in scale and is mainly used to verify the correctness of the operation of reinforcement learning-related functions in the TRL library under a specific model architecture.

Natural Language ProcessingTransformers

AuriStream-1B is a biologically inspired GPT-style autoregressive Transformer model specifically designed to predict cochlear markers in long speech contexts. The model uses a long context window of approximately 20 seconds (4096 tokens) and was trained on the LibriLight dataset (approximately 60,000 hours) for about 500,000 steps. It can learn rich time-aligned representations and generate speech continuations.

Audio ProcessingTransformersEnglish

This model is a specialized psychological counseling model fine-tuned through preference learning based on Llama-3.1-8B-Instruct, excelling in psychological counseling sessions with a win rate surpassing GPT-4o.

Natural Language ProcessingTransformers

A poetry generation model fine-tuned based on GPT-2, trained on the Gutenberg poetry corpus using reinforcement learning, optimized for poetry generation, and performs excellently in style characteristics such as rhyme, coherence, and creativity.

Natural Language ProcessingTransformersEnglish

VITA-1.5 is a multimodal interaction model designed to achieve GPT-4o level real-time vision and voice interaction capabilities.

MultimodalSafetensors

GPepT is a cutting - edge language model based on the GPT - 2 architecture, specifically designed to understand and generate peptide and peptidomimetic sequences. This model contains 738 million parameters and is pre - trained on a curated dataset of ChEMBL bioactive chemical formulas. It can generate diverse peptidomimetic sequences through self - supervised learning, providing an innovative approach for peptide - based therapies.

Scientific ComputingPytorch

A text classification model fine-tuned based on XLM-R Large, specifically designed to identify factual and non-factual statements in German texts. The model uses weakly supervised learning. It is first trained on a Telegram dataset annotated by GPT-4o and then continues to be trained on a manually annotated dataset, achieving an accuracy of 0.9 on the test set.

Natural Language ProcessingTransformersGerman

InternLM-XComposer2.5 is an outstanding text-image understanding and creation model that achieves GPT-4V level performance with just 7B parameters, supporting 24K interleaved text-image context and extendable to 96K long context.

MultimodalTransformers

InternLM-XComposer2.5 is an outstanding image-text understanding and creation model that achieves GPT-4V level capabilities with just 7 billion parameters, supporting long-context window expansion.

MultimodalTransformers

A masked language model based on the diffusion process, similar in scale to GPT2-medium, trained through a forward diffusion process capable of reconstructing original inputs from different masking levels.

Natural Language ProcessingTransformersEnglish

This model is constructed through expert collaboration methods, specializing in mathematical problem solving, capable of autonomously generating solutions or invoking GPT-4 level large models when necessary.

Natural Language ProcessingTransformers

Prometheus 2 is a language model based on Mistral-Instruct, specializing in fine-grained evaluation and reward modeling for Reinforcement Learning from Human Feedback (RLHF), serving as an alternative to GPT-4 evaluation.

Natural Language ProcessingTransformersEnglish

Prometheus 2 is a language model based on Mistral-Instruct, specifically designed for fine-grained evaluation and reinforcement learning from human feedback, serving as an alternative to GPT-4 evaluation.

Natural Language ProcessingTransformersEnglish

Soniox-7B-v1.0 is a powerful large language model built on Mistral 7B, supporting English and code processing with a context length of up to 8K. Its performance in some benchmark tests can be comparable to that of GPT-4, and it has strong problem-solving capabilities.

Natural Language ProcessingTransformers

A GPT2 large model trained on the Anthropic/hh-rlhf helpfulness dataset, specifically designed for helpful response detection or RLHF (Reinforcement Learning from Human Feedback).

Natural Language ProcessingTransformers