Llamacpp quants
This commit is contained in:
ai-modelscope
154
README.md
154
README.md
@@ -1,47 +1,119 @@
|
||||
---
|
||||
license: Apache License 2.0
|
||||
|
||||
#model-type:
|
||||
##如 gpt、phi、llama、chatglm、baichuan 等
|
||||
#- gpt
|
||||
|
||||
#domain:
|
||||
##如 nlp、cv、audio、multi-modal
|
||||
#- nlp
|
||||
|
||||
#language:
|
||||
##语言代码列表 https://help.aliyun.com/document_detail/215387.html?spm=a2c4g.11186623.0.0.9f8d7467kni6Aa
|
||||
#- cn
|
||||
|
||||
#metrics:
|
||||
##如 CIDEr、Blue、ROUGE 等
|
||||
#- CIDEr
|
||||
|
||||
#tags:
|
||||
##各种自定义，包括 pretrained、fine-tuned、instruction-tuned、RL-tuned 等训练方法和其他
|
||||
#- pretrained
|
||||
|
||||
#tools:
|
||||
##如 vllm、fastchat、llamacpp、AdaSeq 等
|
||||
#- vllm
|
||||
base_model:
|
||||
- nitky/Oumuamua-7b-base
|
||||
- nitky/Oumuamua-7b-instruct
|
||||
- tokyotech-llm/Swallow-MS-7b-v0.1
|
||||
- mistralai/Mistral-7B-v0.1
|
||||
- prometheus-eval/prometheus-7b-v2.0
|
||||
- cognitivecomputations/dolphin-2.8-mistral-7b-v02
|
||||
- ZhangShenao/SELM-Zephyr-7B-iter-3
|
||||
- HachiML/Mistral-7B-v0.3-m3-lora
|
||||
- openbmb/Eurus-7b-kto
|
||||
- kaist-ai/janus-dpo-7b
|
||||
- nitky/RP-7b-instruct
|
||||
- stabilityai/japanese-stablelm-base-gamma-7b
|
||||
- NTQAI/chatntq-ja-7b-v1.0
|
||||
- Weyaxi/Einstein-v6-7B
|
||||
- internistai/base-7b-v0.2
|
||||
- ZySec-AI/ZySec-7B
|
||||
library_name: transformers
|
||||
tags:
|
||||
- mergekit
|
||||
- merge
|
||||
language:
|
||||
- ja
|
||||
- en
|
||||
pipeline_tag: text-generation
|
||||
license: apache-2.0
|
||||
quantized_by: bartowski
|
||||
---
|
||||
### 当前模型的贡献者未提供更加详细的模型介绍。模型文件和权重，可浏览“模型文件”页面获取。
|
||||
#### 您可以通过如下git clone命令，或者ModelScope SDK来下载模型
|
||||
|
||||
SDK下载
|
||||
```bash
|
||||
#安装ModelScope
|
||||
pip install modelscope
|
||||
## Llamacpp imatrix Quantizations of Oumuamua-7b-instruct-v2
|
||||
|
||||
Using <a href="https://github.com/ggerganov/llama.cpp/">llama.cpp</a> release <a href="https://github.com/ggerganov/llama.cpp/releases/tag/b3152">b3152</a> for quantization.
|
||||
|
||||
Original model: https://huggingface.co/nitky/Oumuamua-7b-instruct-v2
|
||||
|
||||
All quants made using imatrix option with dataset from [here](https://gist.github.com/bartowski1182/eb213dccb3571f863da82e99418f81e8)
|
||||
|
||||
## Prompt format
|
||||
|
||||
No chat template specified so default is used. This may be incorrect, check original model card for details.
|
||||
|
||||
```
|
||||
```python
|
||||
#SDK模型下载
|
||||
from modelscope import snapshot_download
|
||||
model_dir = snapshot_download('bartowski/Oumuamua-7b-instruct-v2-GGUF')
|
||||
```
|
||||
Git下载
|
||||
```
|
||||
#Git模型下载
|
||||
git clone https://www.modelscope.cn/bartowski/Oumuamua-7b-instruct-v2-GGUF.git
|
||||
<s> [INST] <<SYS>>
|
||||
{system_prompt}
|
||||
<</SYS>>
|
||||
|
||||
{prompt} [/INST] </s>
|
||||
```
|
||||
|
||||
<p style="color: lightgrey;">如果您是本模型的贡献者，我们邀请您根据<a href="https://modelscope.cn/docs/ModelScope%E6%A8%A1%E5%9E%8B%E6%8E%A5%E5%85%A5%E6%B5%81%E7%A8%8B%E6%A6%82%E8%A7%88" style="color: lightgrey; text-decoration: underline;">模型贡献文档</a>，及时完善模型卡片内容。</p>
|
||||
## Download a file (not the whole branch) from below:
|
||||
|
||||
| Filename | Quant type | File Size | Description |
|
||||
| -------- | ---------- | --------- | ----------- |
|
||||
| [Oumuamua-7b-instruct-v2-Q8_0.gguf](https://huggingface.co/bartowski/Oumuamua-7b-instruct-v2-GGUF/blob/main/Oumuamua-7b-instruct-v2-Q8_0.gguf) | Q8_0 | 8.11GB | Extremely high quality, generally unneeded but max available quant. |
|
||||
| [Oumuamua-7b-instruct-v2-Q6_K.gguf](https://huggingface.co/bartowski/Oumuamua-7b-instruct-v2-GGUF/blob/main/Oumuamua-7b-instruct-v2-Q6_K.gguf) | Q6_K | 6.42GB | Very high quality, near perfect, *recommended*. |
|
||||
| [Oumuamua-7b-instruct-v2-Q5_K_M.gguf](https://huggingface.co/bartowski/Oumuamua-7b-instruct-v2-GGUF/blob/main/Oumuamua-7b-instruct-v2-Q5_K_M.gguf) | Q5_K_M | 5.63GB | High quality, *recommended*. |
|
||||
| [Oumuamua-7b-instruct-v2-Q5_K_S.gguf](https://huggingface.co/bartowski/Oumuamua-7b-instruct-v2-GGUF/blob/main/Oumuamua-7b-instruct-v2-Q5_K_S.gguf) | Q5_K_S | 5.50GB | High quality, *recommended*. |
|
||||
| [Oumuamua-7b-instruct-v2-Q4_K_M.gguf](https://huggingface.co/bartowski/Oumuamua-7b-instruct-v2-GGUF/blob/main/Oumuamua-7b-instruct-v2-Q4_K_M.gguf) | Q4_K_M | 4.88GB | Good quality, uses about 4.83 bits per weight, *recommended*. |
|
||||
| [Oumuamua-7b-instruct-v2-Q4_K_S.gguf](https://huggingface.co/bartowski/Oumuamua-7b-instruct-v2-GGUF/blob/main/Oumuamua-7b-instruct-v2-Q4_K_S.gguf) | Q4_K_S | 4.66GB | Slightly lower quality with more space savings, *recommended*. |
|
||||
| [Oumuamua-7b-instruct-v2-IQ4_XS.gguf](https://huggingface.co/bartowski/Oumuamua-7b-instruct-v2-GGUF/blob/main/Oumuamua-7b-instruct-v2-IQ4_XS.gguf) | IQ4_XS | 4.43GB | Decent quality, smaller than Q4_K_S with similar performance, *recommended*. |
|
||||
| [Oumuamua-7b-instruct-v2-Q3_K_L.gguf](https://huggingface.co/bartowski/Oumuamua-7b-instruct-v2-GGUF/blob/main/Oumuamua-7b-instruct-v2-Q3_K_L.gguf) | Q3_K_L | 4.35GB | Lower quality but usable, good for low RAM availability. |
|
||||
| [Oumuamua-7b-instruct-v2-Q3_K_M.gguf](https://huggingface.co/bartowski/Oumuamua-7b-instruct-v2-GGUF/blob/main/Oumuamua-7b-instruct-v2-Q3_K_M.gguf) | Q3_K_M | 4.05GB | Even lower quality. |
|
||||
| [Oumuamua-7b-instruct-v2-IQ3_M.gguf](https://huggingface.co/bartowski/Oumuamua-7b-instruct-v2-GGUF/blob/main/Oumuamua-7b-instruct-v2-IQ3_M.gguf) | IQ3_M | 3.82GB | Medium-low quality, new method with decent performance comparable to Q3_K_M. |
|
||||
| [Oumuamua-7b-instruct-v2-Q3_K_S.gguf](https://huggingface.co/bartowski/Oumuamua-7b-instruct-v2-GGUF/blob/main/Oumuamua-7b-instruct-v2-Q3_K_S.gguf) | Q3_K_S | 3.70GB | Low quality, not recommended. |
|
||||
| [Oumuamua-7b-instruct-v2-IQ3_XS.gguf](https://huggingface.co/bartowski/Oumuamua-7b-instruct-v2-GGUF/blob/main/Oumuamua-7b-instruct-v2-IQ3_XS.gguf) | IQ3_XS | 3.55GB | Lower quality, new method with decent performance, slightly better than Q3_K_S. |
|
||||
| [Oumuamua-7b-instruct-v2-IQ3_XXS.gguf](https://huggingface.co/bartowski/Oumuamua-7b-instruct-v2-GGUF/blob/main/Oumuamua-7b-instruct-v2-IQ3_XXS.gguf) | IQ3_XXS | 3.38GB | Lower quality, new method with decent performance, comparable to Q3 quants. |
|
||||
| [Oumuamua-7b-instruct-v2-Q2_K.gguf](https://huggingface.co/bartowski/Oumuamua-7b-instruct-v2-GGUF/blob/main/Oumuamua-7b-instruct-v2-Q2_K.gguf) | Q2_K | 3.27GB | Very low quality but surprisingly usable. |
|
||||
| [Oumuamua-7b-instruct-v2-IQ2_M.gguf](https://huggingface.co/bartowski/Oumuamua-7b-instruct-v2-GGUF/blob/main/Oumuamua-7b-instruct-v2-IQ2_M.gguf) | IQ2_M | 3.05GB | Very low quality, uses SOTA techniques to also be surprisingly usable. |
|
||||
| [Oumuamua-7b-instruct-v2-IQ2_S.gguf](https://huggingface.co/bartowski/Oumuamua-7b-instruct-v2-GGUF/blob/main/Oumuamua-7b-instruct-v2-IQ2_S.gguf) | IQ2_S | 2.86GB | Very low quality, uses SOTA techniques to be usable. |
|
||||
| [Oumuamua-7b-instruct-v2-IQ2_XS.gguf](https://huggingface.co/bartowski/Oumuamua-7b-instruct-v2-GGUF/blob/main/Oumuamua-7b-instruct-v2-IQ2_XS.gguf) | IQ2_XS | 2.76GB | Very low quality, uses SOTA techniques to be usable. |
|
||||
|
||||
## Downloading using huggingface-cli
|
||||
|
||||
First, make sure you have hugginface-cli installed:
|
||||
|
||||
```
|
||||
pip install -U "huggingface_hub[cli]"
|
||||
```
|
||||
|
||||
Then, you can target the specific file you want:
|
||||
|
||||
```
|
||||
huggingface-cli download bartowski/Oumuamua-7b-instruct-v2-GGUF --include "Oumuamua-7b-instruct-v2-Q4_K_M.gguf" --local-dir ./
|
||||
```
|
||||
|
||||
If the model is bigger than 50GB, it will have been split into multiple files. In order to download them all to a local folder, run:
|
||||
|
||||
```
|
||||
huggingface-cli download bartowski/Oumuamua-7b-instruct-v2-GGUF --include "Oumuamua-7b-instruct-v2-Q8_0.gguf/*" --local-dir Oumuamua-7b-instruct-v2-Q8_0
|
||||
```
|
||||
|
||||
You can either specify a new local-dir (Oumuamua-7b-instruct-v2-Q8_0) or download them all in place (./)
|
||||
|
||||
## Which file should I choose?
|
||||
|
||||
A great write up with charts showing various performances is provided by Artefact2 [here](https://gist.github.com/Artefact2/b5f810600771265fc1e39442288e8ec9)
|
||||
|
||||
The first thing to figure out is how big a model you can run. To do this, you'll need to figure out how much RAM and/or VRAM you have.
|
||||
|
||||
If you want your model running as FAST as possible, you'll want to fit the whole thing on your GPU's VRAM. Aim for a quant with a file size 1-2GB smaller than your GPU's total VRAM.
|
||||
|
||||
If you want the absolute maximum quality, add both your system RAM and your GPU's VRAM together, then similarly grab a quant with a file size 1-2GB Smaller than that total.
|
||||
|
||||
Next, you'll need to decide if you want to use an 'I-quant' or a 'K-quant'.
|
||||
|
||||
If you don't want to think too much, grab one of the K-quants. These are in format 'QX_K_X', like Q5_K_M.
|
||||
|
||||
If you want to get more into the weeds, you can check out this extremely useful feature chart:
|
||||
|
||||
[llama.cpp feature matrix](https://github.com/ggerganov/llama.cpp/wiki/Feature-matrix)
|
||||
|
||||
But basically, if you're aiming for below Q4, and you're running cuBLAS (Nvidia) or rocBLAS (AMD), you should look towards the I-quants. These are in format IQX_X, like IQ3_M. These are newer and offer better performance for their size.
|
||||
|
||||
These I-quants can also be used on CPU and Apple Metal, but will be slower than their K-quant equivalent, so speed vs performance is a tradeoff you'll have to decide.
|
||||
|
||||
The I-quants are *not* compatible with Vulcan, which is also AMD, so if you have an AMD card double check if you're using the rocBLAS build or the Vulcan build. At the time of writing this, LM Studio has a preview with ROCm support, and other inference engines have specific builds for ROCm.
|
||||
|
||||
Want to support my work? Visit my ko-fi page here: https://ko-fi.com/bartowski
|
||||
|
||||
Reference in New Issue
Block a user