Block Entropy Models

Block Entropy models are open source LLMs that we are currently hosting.

Block Entropy Base Models

Model: be-405b-instruct-llama3.1

This is a 4.5bpw quantization of the largest model, Llama 3.1 405B. The max context window length at the moment is 16k. This is in beta, experimental stage.

Model: be-72b-chat-qwen2

This is a 8.0bpw quantization of the Qwen2 72B parameter model. It has 16K context, 8.0bpw

Model: be-70b-instruct-llama3.1

This is a 6.0bpw of the Llama 3.1 70B model, supporting 128K context.

Model: be-70b-instruct-llama3

This is a 6.0bpw quantization of the Llama 3.0 70B parameter model at 8K context.

Block Entropy Community LoRAs or Merge Models

Model: be-120b-tessxl

This is a 4.5bpw quantization of the Tess XL model, a frankenmerge of Llama and a fine tune of goliath. It has a 12K context via NTK Rope.

Model: be-120b-goliath

This is a low rank adapter of the Goliath 120B model. It runs off the base TessXL model as a rank 64 adapter. 12K context via NTK Rope.

Model: be-120b-miqu

This is a dynamic layering of the Miqu 70B parameter model to enhance the model to 120B. It contains 8K context, at 5.0bpw

Model: be-120b-miquliz

This is a low rank adapter of the Miquliz model on top of the 120B Miqu model. While this has a higher rank (256 LoRA), this an approximation of the 120B Miquliz at 8K context, 5.0bpw.

Model: be-72b-magnumv1-qwen2

This is a 8.0bpw quantization of the Qwen2 72B parameter model. It has 8K context, 8.0bpw

Model: be-70b-cat-llama3

This is a 64 rank approximation of the Cat-llama3-instruct model at 6.0bpw.

Model: be-70b-openbiollm-llama3

This is a 128 rank approximation of the OpenBioLLM model at 8k context, 6.0bpw.

Model: be-70b-lumimaid-llama3

This is a 64 rank approximation of the Lumimaid Llama 3 model at 6.0bpw.

Model: be-70b-smaug-llama3-abliterated

This is a 64 rank approximation of the Smaug Abliterated Llama 3 model at 8K context, 6.0bpw.