models(gallery): add rombos-llm-v2.5.1-qwen-3b (#3778)

Signed-off-by: Ettore Di Giacinto <mudler@localai.io>
This commit is contained in:
Ettore Di Giacinto 2024-10-10 10:47:41 +02:00 committed by GitHub
parent 85a3cc8d8f
commit f41965bfb5
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194

View File

@ -396,6 +396,28 @@
- filename: T.E-8.1-Q4_K_M-imat.gguf
sha256: 1b7892b82c01ea4cbebe34cd00f9836cbbc369fc3247c1f44a92842201e7ec0b
uri: huggingface://Lewdiculous/T.E-8.1-GGUF-IQ-Imatrix-Request/T.E-8.1-Q4_K_M-imat.gguf
- !!merge <<: *qwen25
name: "rombos-llm-v2.5.1-qwen-3b"
icon: https://cdn-uploads.huggingface.co/production/uploads/642cc1c253e76b4c2286c58e/pNDtgE5FDkxxvbG4qiZ1A.jpeg
urls:
- https://huggingface.co/QuantFactory/Rombos-LLM-V2.5.1-Qwen-3b-GGUF
description: |
Rombos-LLM-V2.5.1-Qwen-3b is a little experiment that merges a high-quality LLM, arcee-ai/raspberry-3B, using the last step of the Continuous Finetuning method outlined in a Google document. The merge is done using the mergekit with the following parameters:
- Models: Qwen2.5-3B-Instruct, raspberry-3B
- Merge method: ties
- Base model: Qwen2.5-3B
- Parameters: weight=1, density=1, normalize=true, int8_mask=true
- Dtype: bfloat16
The model has been evaluated on various tasks and datasets, and the results are available on the Open LLM Leaderboard. The model has shown promising performance across different benchmarks.
overrides:
parameters:
model: Rombos-LLM-V2.5.1-Qwen-3b.Q4_K_M.gguf
files:
- filename: Rombos-LLM-V2.5.1-Qwen-3b.Q4_K_M.gguf
sha256: 656c342a2921cac8912e0123fc295c3bb3d631a85c671c12a3843a957e46d30d
uri: huggingface://QuantFactory/Rombos-LLM-V2.5.1-Qwen-3b-GGUF/Rombos-LLM-V2.5.1-Qwen-3b.Q4_K_M.gguf
- &archfunct
license: apache-2.0
tags: