基于ExLlamaV2 v0.0.15量化的文本生成模型,提供8.0/6.5/5.0等多分支选择,平衡性能与显存占用,适用于不同配置设备。【此简介由AI生成】
以下内容由 AI 翻译,如有问题请 点此提交 issue 反馈
license: cc-by-nc-4.0 library_name: transformers tags:
- mergekit
- merge base_model:
- Sao10K/Fimbulvetr-10.7B-v1
- saishf/Kuro-Lotus-10.7B model-index:
- name: Fimbulvetr-Kuro-Lotus-10.7B
results:
- task:
type: text-generation
name: Text Generation
dataset:
name: AI2 Reasoning Challenge (25-Shot)
type: ai2_arc
config: ARC-Challenge
split: test
args:
num_few_shot: 25
metrics:
- type: acc_norm value: 69.54 name: normalized accuracy source: url: https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=saishf/Fimbulvetr-Kuro-Lotus-10.7B name: Open LLM Leaderboard
- task:
type: text-generation
name: Text Generation
dataset:
name: HellaSwag (10-Shot)
type: hellaswag
split: validation
args:
num_few_shot: 10
metrics:
- type: acc_norm value: 87.87 name: normalized accuracy source: url: https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=saishf/Fimbulvetr-Kuro-Lotus-10.7B name: Open LLM Leaderboard
- task:
type: text-generation
name: Text Generation
dataset:
name: MMLU (5-Shot)
type: cais/mmlu
config: all
split: test
args:
num_few_shot: 5
metrics:
- type: acc value: 66.99 name: accuracy source: url: https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=saishf/Fimbulvetr-Kuro-Lotus-10.7B name: Open LLM Leaderboard
- task:
type: text-generation
name: Text Generation
dataset:
name: TruthfulQA (0-shot)
type: truthful_qa
config: multiple_choice
split: validation
args:
num_few_shot: 0
metrics:
- type: mc2 value: 60.95 source: url: https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=saishf/Fimbulvetr-Kuro-Lotus-10.7B name: Open LLM Leaderboard
- task:
type: text-generation
name: Text Generation
dataset:
name: Winogrande (5-shot)
type: winogrande
config: winogrande_xl
split: validation
args:
num_few_shot: 5
metrics:
- type: acc value: 84.14 name: accuracy source: url: https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=saishf/Fimbulvetr-Kuro-Lotus-10.7B name: Open LLM Leaderboard
- task:
type: text-generation
name: Text Generation
dataset:
name: GSM8k (5-shot)
type: gsm8k
config: main
split: test
args:
num_few_shot: 5
metrics:
- type: acc value: 66.87 name: accuracy source: url: https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=saishf/Fimbulvetr-Kuro-Lotus-10.7B name: Open LLM Leaderboard quantized_by: bartowski pipeline_tag: text-generation
- task:
type: text-generation
name: Text Generation
dataset:
name: AI2 Reasoning Challenge (25-Shot)
type: ai2_arc
config: ARC-Challenge
split: test
args:
num_few_shot: 25
metrics:
Fimbulvetr-Kuro-Lotus-10.7B 的 Exllama v2 量化版本
使用 turboderp 的 ExLlamaV2 v0.0.15 进行量化。
请注意:"main"分支仅包含 measurement.json 文件,请下载其他分支以获取完整模型(详见下方)
每个分支对应不同的权重比特精度,主分支仅包含用于进一步转换的 measurement.json 文件。
原始模型:https://huggingface.co/saishf/Fimbulvetr-Kuro-Lotus-10.7B
| 分支 | 权重比特 | lm_head 比特 | VRAM (4k) | VRAM (16k) | VRAM (32k) | 描述 |
|---|---|---|---|---|---|---|
| 8_0 | 8.0 | 8.0 | 11.9 GB | 13.3 GB | 15.3 GB | ExLlamaV2 可实现的最高质量,接近未量化性能 |
| 6_5 | 6.5 | 8.0 | 10.3 GB | 11.7 GB | 13.7 GB | 与8.0版本高度接近,尺寸与性能的最佳平衡,推荐使用 |
| 5_0 | 5.0 | 6.0 | 8.3 GB | 9.7 GB | 11.7 GB | 质量略低于6.5版本,但可在8GB显存设备上运行 |
| 4_25 | 4.25 | 6.0 | 7.4 GB | 8.6 GB | 10.6 GB | 等效GPTQ权重比特精度,质量略优 |
| 3_5 | 3.5 | 6.0 | 6.4 GB | 7.8 GB | 9.8 GB | 较低质量版本,仅建议必要时使用 |
下载说明
使用 git:
git clone --single-branch --branch 6_5 https://huggingface.co/bartowski/Fimbulvetr-Kuro-Lotus-10.7B-exl2 Fimbulvetr-Kuro-Lotus-10.7B-exl2-6_5
借助 Hugging Face Hub(感谢 TheBloke 提供的指南):
pip3 install huggingface-hub
要将 main 分支(仅适用于仅关注 measurement.json 的情况)下载到名为 Fimbulvetr-Kuro-Lotus-10.7B-exl2 的文件夹中:
mkdir Fimbulvetr-Kuro-Lotus-10.7B-exl2
huggingface-cli download bartowski/Fimbulvetr-Kuro-Lotus-10.7B-exl2 --local-dir Fimbulvetr-Kuro-Lotus-10.7B-exl2 --local-dir-use-symlinks False
要从其他分支下载,请添加 --revision 参数:
Linux:
mkdir Fimbulvetr-Kuro-Lotus-10.7B-exl2-6_5
huggingface-cli download bartowski/Fimbulvetr-Kuro-Lotus-10.7B-exl2 --revision 6_5 --local-dir Fimbulvetr-Kuro-Lotus-10.7B-exl2-6_5 --local-dir-use-symlinks False
Windows(有时似乎不支持下划线在文件夹名称中使用?):
mkdir Fimbulvetr-Kuro-Lotus-10.7B-exl2-6.5
huggingface-cli download bartowski/Fimbulvetr-Kuro-Lotus-10.7B-exl2 --revision 6_5 --local-dir Fimbulvetr-Kuro-Lotus-10.7B-exl2-6.5 --local-dir-use-symlinks False
想支持我的工作吗?请访问我的 Ko-fi 页面:https://ko-fi.com/bartowski