Exllama v2 Quantizations of NeuralBeagle14-7B
Using turboderp's ExLlamaV2 v0.0.11 for quantization.
The "main" branch only contains the measurement.json, download one of the other branches for the model (see below)
Each branch contains an individual bits per weight, with the main one containing only the meaurement.json for further conversions.
Original model: https://huggingface.co./mlabonne/NeuralBeagle14-7B
Model Size: 7b
| Branch | Bits | lm_head bits | Dataset | Size | Description |
|---|---|---|---|---|---|
| 8_0 | 8.0 | 8.0 | Default | 9.8 GB | Maximum quality that ExLlamaV2 can produce, near unquantized performance. |
| 6_5 | 6.5 | 8.0 | Default | 8.6 GB | Very similar to 8.0, good tradeoff of size vs performance, recommended. |
| 5_0 | 5.0 | 6.0 | Default | 7.4 GB | Slightly lower perplexity vs 6.5. |
| 4_0 | 4.0 | 6.0 | Default | 6.5 GB | Just under GPTQ equivalent bits per weight. |
| 3_5 | 3.5 | 6.0 | Default | 6.1 GB | Lower quality, only use if you have to. |
All VRAM requirements estimated from 16k context. For 32k context add ~2 GB.
Download instructions
With git:
git clone --single-branch --branch 4_0 https://huggingface.co./bartowski/NeuralBeagle14-7B-exl2
With huggingface hub (credit to TheBloke for instructions):
pip3 install huggingface-hub
To download the main (only useful if you only care about measurement.json) branch to a folder called NeuralBeagle14-7B-exl2:
mkdir NeuralBeagle14-7B-exl2
huggingface-cli download bartowski/NeuralBeagle14-7B-exl2 --local-dir NeuralBeagle14-7B-exl2 --local-dir-use-symlinks False
To download from a different branch, add the --revision parameter:
mkdir NeuralBeagle14-7B-exl2-6_5
huggingface-cli download bartowski/NeuralBeagle14-7B-exl2 --revision 6_5 --local-dir NeuralBeagle14-7B-exl2-6_5 --local-dir-use-symlinks False