NVIDIA Ever-Soupcocs OP Open Code Code Reasoning Models (32b, 14b, 7b)

Unvidia continues to press the borders of an open AI for opening Open Code Recing (OCR) Model Suite – High-Performance Trio Long-Performance TRIO Languages ​​that apply to Code consultation and solving problems. 32b, 14b, and 7b variants, all were issued under the Apache 2.0 License.

Considered the best beat

This page Open Code of consultation (OCR) The models came with The achievement of ASSSNEN BETCHMARK EXENIONSoutput O3-mini of O3-mini and O1 (low) models in LiveCodebelch Benchmark. LiveCodbench is a comprehensive program of testing code of action that reflects the abuse, withdrawing code, and the termination of psychiatry in the world's realmillers in the world. Direct comparisons, OC2 Kanvia 32b's OCR model reaches the boxing boards to open power models.

This work jump is said to be reported only in a building model, but to Lvidi's Culture “OCR Dataset” – High Quality Corpus, Code Centric Corpus is designed to emphasize teaching, consultation, and resolving problems of different code code issues. According to Nvidia, this results in 30% upgrading to the efficiency of the TokenTo allow models that generate an accurate code and reasonable effects with a few tokens.

A model line of all the use of the use

Open Code that consults suite is entering Three scales of parameter:

• Opencoderoaving-Nemotron-32b
• Opencoderoing-Nemotron-14b
• OPENCODEROINAING-NEMOTRO-7B

Individual ratings by working. A unique 32b submits the results of the State State to operate effectiveness and research; The 14B model provides strong-related skills for reduced computer needs, and is different 7B ready to oppressed resources while storing competition in competition.

All models are trained using Nemotn BuildingThe Backbida based on nvidia is designed for many languages, many jobs. The weight of the model and configuration is available at the National System:

It is compatible with an open carefully

An important feature of these types Box Account With the famous wooden wood:

• llama.cpp In order to obtain the lack of a lack of CPU / GPU
• vLLM With well-functioning gpu and imaginary decorations
• Transformers By tainting the training and testing of pipes
• TGI (Fatilitization) API Shipment

This fluctuations allow developers, businesses and businesses to connect these models to Code Code Code or AI with less than infinity.

The step forward to drop of open code

This release, unvidia giveth heavily on the growing ecosystem of open code models. Direct Code Code – Base Controlled by Models – and we are released under open open and approved license, Invidia gives the broader community of AI and developers to build, advanced thought models in production.

The open consultation code for SUITE adds to the growing portfolio of open LLMs and strengthens its status in an accessible, an obvious development of AI. Even if you create the developer copies, the default copies of the default codes, or code generation services, these models provide an alternative, very expensive, social, and public.

Look at 32b model, 14b model, 7B model and 32b of teaching orders – different. Also, don't forget to follow Sane.

Here is a short opinion of what we build in MarktechPost: