vrr/kvcache-ai-ktransformers
2
0
Fork 0
mirror of https://github.com/kvcache-ai/ktransformers.git synced 2026-04-28 20:00:06 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions 4

[docs]: refine dpo tutorial (#1739)
Some checks failed
Deploy / deploy (macos-latest) (push) Has been cancelled
Details
Deploy / deploy (ubuntu-latest) (push) Has been cancelled
Details
Book-CI / test-1 (push) Has been cancelled
Details
Book-CI / test-2 (push) Has been cancelled
Details
Deploy / deploy (windows-latest) (push) Has been cancelled
Details
Book-CI / test (push) Has been cancelled
Details

Browse source
This commit is contained in:
mrhaoxx 2025-12-22 18:44:24 +08:00 • committed by GitHub
parent 0bce173e3b
commit dee1e211d5
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
1 changed files with 4 additions and 4 deletions
Download patch file Download diff file Expand all files Collapse all files

8
doc/en/DPO_tutorial.md
View file

@ -61,7 +61,7 @@ pip install custom_flashinfer/
## Prepare Models
We uses `DeepSeek-V2-Lite-Chat` as an example here. You can replace it with other models such as Kimi K2.
We uses `deepseek-ai/DeepSeek-V2-Lite` as an example here. You can replace it with other models such as Kimi K2.
## How to start
@ -80,7 +80,7 @@ For example, we provide the YAML file as follows:
```YAML
### model
model_name_or_path: DeepSeek-V2-Lite-Chat
model_name_or_path: deepseek-ai/DeepSeek-V2-Lite
trust_remote_code: true
### method
@ -114,7 +114,7 @@ report_to: none # choices: [none, wandb, tensorboard, swanlab, mlflow]
per_device_train_batch_size: 1
gradient_accumulation_steps: 8
learning_rate: 5.0e-6
num_train_epochs: 0.1
num_train_epochs: 3
lr_scheduler_type: cosine
warmup_ratio: 0.1
bf16: true
@ -130,7 +130,7 @@ chunk_size: 8192
For more details about --kt_optimize_rule, please refer to https://github.com/kvcache-ai/ktransformers/blob/main/doc/en/KTransformers-Fine-Tuning_User-Guide.md
（2）examples/inference/deepseek2_lora_dpo_kt.yaml
Then, you can use the lora adapter saved in `saves/Kllama_deepseekV2_DPO` for inference the same as the sft training. For example,
```YAML
model_name_or_path: DeepSeek-V2-Lite-Chat