* evals calculated with llama.cpp llama-perplexity

tiiuae/Falcon3-7B-Instruct neopolitized with projected shards and fragments of Qwen/Qwen2.5-7B-Instruct.

projection method: 0
merge method: 1
layers: 0-27 [x->x]
alpha: 0.8-0.9
tensors: emb, head, attn_q, attn_k, attn_v, attn_o, ffn_g, ffn_u, ffn_d

                             8 w  w       
8d8b. .d88b .d8b. 88b. .d8b. 8 w w8ww .d88
8P Y8 8.dP' 8' .8 8  8 8' .8 8 8  8   8  8
8   8 `Y88P `Y8P' 88P' `Y8P' 8 8  Y8P `Y88
                  8

Downloads last month: 21

GGUF

Model size

7B params

Architecture

llama

Hardware compatibility

16-bit

Inference Providers NEW

This model isn't deployed by any Inference Provider. 🙋 Ask for provider support

Model tree for neopolita/Neo-Falcon3-7B-Instruct-v0

Base model

tiiuae/Falcon3-7B-Base

Finetuned

tiiuae/Falcon3-7B-Instruct

Quantized

(52)

this model

Collection including neopolita/Neo-Falcon3-7B-Instruct-v0

Neo

Collection

Neo is a collection of model merges using a novel heterogeneous LLM merging technique (paper in progress). • 5 items • Updated Jan 5 • 1