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!
- Downloads last month
- 11,080
Model tree for lastmass/llama3.2-chinese
Base model
meta-llama/Llama-3.2-3B