continued pretraining with wikipedia-zh

dataset = load_dataset("wikimedia/wikipedia", "20231101.zh", split = "train",)

and sft with FreedomIntelligence/alpaca-gpt4-zh

alpaca_dataset = load_dataset("FreedomIntelligence/alpaca-gpt4-zh", split = "train")

using unsloth to fine-tuning the model

use the model-unsloth.gguf file or model-unsloth-Q4_K_M.gguf file in llama.cpp or a UI based system like GPT4All.

aslo you can make the ollama model with modelfile

template

以下是描述一项任务的指令。请输出合适的内容回应指令。
### Input:
大象和猎豹的奔跑速度谁更快,简单说明原因.

### Response:
猎豹的奔跑速度比大象快得多。

猎豹:是世界上奔跑速度最快的陆地动物之一,短距离内可以达到约 100-120 公里/小时(约 60-75 英里/小时)。
大象:虽然大象体型巨大,但它们的速度较慢,奔跑速度最高约为 40 公里/小时(约 25 英里/小时)。
因此,猎豹在速度上远远超过了大象。

### Input:
鱼为什么能在水里呼吸。

### Response:
鱼能够在水中呼吸,主要是因为它们有一种特殊的呼吸器官——鳃。鳃能够从水中提取氧气,并排出二氧化碳,这个过程使鱼能够在水中生存。
{{ if .Prompt }}

### 指令:
{{ .Prompt }}

{{ end }}### 回应:
{{ .Response }}<|end_of_text|>

or just use ollama run lastmass/llama3.2-chinese

Community Feedback Request

Dear users who have downloaded this model,

I've noticed a significant increase in downloads recently, which I truly appreciate. If possible, I would be grateful if you could share your experience and feedback in our community. Your insights would be invaluable for future improvements.

Thank you for your support!


also see https://huggingface.co/lastmass/llama3.1-Medical-Assistant

Downloads last month
9,021
GGUF
Model size
3.61B params
Architecture
llama

4-bit

16-bit

Inference API
Unable to determine this model's library. Check the docs .

Model tree for lastmass/llama3.2-chinese

Quantized
(44)
this model

Dataset used to train lastmass/llama3.2-chinese