|Published several new, ground-breaking models|
- Published General-English-Image-Caption-Blip-2, a scalable multimodal pre-training method that enables any Large Language Models (LLMs) to ingest and understand images. It unlocks the capabilities of zero-shot image-to-text generation.
- Published Falcon-7B-Instruct, a 7B parameters LLM based on Falcon-7B and fine-tuned on instructions and conversational data; they thus lend better to popular assistant-style tasks.
- Published Hkunlp_Instructor-XL, an embedding model that can generate text embeddings tailored to any task (e.g., classification, clustering, text evaluation, etc.) and domains (e.g., science, finance, etc.) by simply providing the task instruction, without any fine-tuning.
- Published Llama2-7B-Chat, a fine-tuned LLM from Meta that is optimized for dialogue use cases.
- Published Llama2-13B-Chat, a fine-tuned LLM from Meta that is optimized for dialogue use cases.
- Published Text-Bison, a foundation model from GCP (Google Cloud Platform) that is optimized for a variety of natural language tasks such as sentiment analysis, entity extraction, content creation, document summaries, answers to questions, and labels that classify content.
- Published Code-Gecko, a foundation model from GCP that supports code completion. It generates new code based on the code that was recently typed by a user.
- Published Code-Bison, a foundation model from GCP that supports code generation. It generates code based on a natural language description. For example, it can create functions, web pages, and unit tests.
- Published Textembedding-Gecko, an embedding model from GCP that generates embeddings from the given text, which can be used for different language-related tasks.
- Published Detr-Resnet-101, a DEtection TRansformer (DETR) object detection model that is trained end-to-end on COCO 2017 dataset (118k annotated images).
- Published General-Image-Recognition-Deit-Base, a Data-Efficient Image Transformer (DeiT) image classification model that is pre-trained and fine-tuned on ImageNet-1k (1 million images, 1,000 classes).
- Published Claude-v2, a chat completion model from Anthropic, driven by an LLM, for generating contextually relevant and coherent responses.
- Published General-Image-Recognition-Deit-Base, a Data-Efficient Image Transformer (DeiT), state-of-the-art image classification model that is pre-trained and fine-tuned on ImageNet-1k (1 million images, 1,000 classes).
- Published General-English-Image-Caption-Blip-2-6_7B, a state-of-the-art image captioning model with 6.7B parameters.
- Published Multimodal-Embedder-Blip-2 , a scalable multimodal pre-training method that enables any LLMs to ingest and understand images. It unlocks the capabilities of zero-shot image-to-text generation.
- Published XGen-7B-8K-Instruct, a powerful 7-billion parameter LLM trained on up to 8K sequence length with fine-tuning on instructional data, enabling robust long sequence.
- Published MPT-Instruct-7B, a decoder-style transformer LLM, fine-tuned for efficient short-form instruction with 6.7B parameters.
|Added ability to customize Hugging Face and MMCV (OpenMMLab Computer Vision) deep training templates using the Python config file format|
- You can now add your own custom model configuration when creating a text classification model using the Hugging Face deep training template.
- You can also add custom configurations to MMClassification, MMDetection, and MMSegmentation deep training templates. You can customize their loss functions, backbones, necks, heads, and more.
|Fixed an issue that caused the model evaluation process to break when numerous concepts were used|
- Model evaluation now works as desired.
|Fixed an issue with the A21 Jurassic generative model that caused it to cache output prompts, resulting in repetitive responses upon subsequent usage|
- The A21 Jurassic model now generates new responses, providing different outputs each time the page is refreshed.
|Fixed an issue where models and workflows ignored new app and user IDs|
- Previously, any attempts to rename an app or a user ID, or to relocate the app to an organization (equivalent to altering the user ID), resulted in the models and workflows failing to recognize these updated values. We fixed the issue.
|Fixed an issue where it was not possible to update a model's visibility|
- Previously, if you edited a model's visibility and published the changes, trying to edit the model's visibility again could not work. We fixed the issue.
|Developer Preview with Request-Only Access||Added ability to import models from Hugging Face|
- You can now import models with permissive licenses from Hugging Face and use them on the Clarifai platform.
|Developer Preview with Request-Only Access||Added ability to fine-tune text-to-text models|
- Advanced model builders can now further customize the behavior and output of the text-to-text models for specific text generation tasks. They can train the models on specific datasets to adapt their behavior for particular tasks or domains.