add vllm config

Former-commit-id: 95365f0ce4f362bde7de8b679b54b548d7055bfb
2024-11-10 21:28:18 +08:00
parent fcb6283a72
commit 1e6f96508a
34 changed files with 44 additions and 34 deletions
--- a/examples/train_lora/llama3_lora_dpo.yaml
+++ b/examples/train_lora/llama3_lora_dpo.yaml
@@ -12,7 +12,7 @@ pref_loss: sigmoid  # choices: [sigmoid (dpo), orpo, simpo]
 ### dataset
 dataset: dpo_en_demo
 template: llama3
-cutoff_len: 1024
+cutoff_len: 2048
 max_samples: 1000
 overwrite_cache: true
 preprocessing_num_workers: 16
--- a/examples/train_lora/llama3_lora_kto.yaml
+++ b/examples/train_lora/llama3_lora_kto.yaml
@@ -11,7 +11,7 @@ pref_beta: 0.1
 ### dataset
 dataset: kto_en_demo
 template: llama3
-cutoff_len: 1024
+cutoff_len: 2048
 max_samples: 1000
 overwrite_cache: true
 preprocessing_num_workers: 16
--- a/examples/train_lora/llama3_lora_ppo.yaml
+++ b/examples/train_lora/llama3_lora_ppo.yaml
@@ -11,7 +11,7 @@ lora_target: all
 ### dataset
 dataset: identity,alpaca_en_demo
 template: llama3
-cutoff_len: 1024
+cutoff_len: 2048
 max_samples: 1000
 overwrite_cache: true
 preprocessing_num_workers: 16
--- a/examples/train_lora/llama3_lora_predict.yaml
+++ b/examples/train_lora/llama3_lora_predict.yaml
@@ -10,7 +10,7 @@ finetuning_type: lora
 ### dataset
 eval_dataset: identity,alpaca_en_demo
 template: llama3
-cutoff_len: 1024
+cutoff_len: 2048
 max_samples: 50
 overwrite_cache: true
 preprocessing_num_workers: 16
--- a/examples/train_lora/llama3_lora_pretrain.yaml
+++ b/examples/train_lora/llama3_lora_pretrain.yaml
@@ -9,7 +9,7 @@ lora_target: all

 ### dataset
 dataset: c4_demo
-cutoff_len: 1024
+cutoff_len: 2048
 max_samples: 1000
 overwrite_cache: true
 preprocessing_num_workers: 16
--- a/examples/train_lora/llama3_lora_reward.yaml
+++ b/examples/train_lora/llama3_lora_reward.yaml
@@ -10,7 +10,7 @@ lora_target: all
 ### dataset
 dataset: dpo_en_demo
 template: llama3
-cutoff_len: 1024
+cutoff_len: 2048
 max_samples: 1000
 overwrite_cache: true
 preprocessing_num_workers: 16
--- a/examples/train_lora/llama3_lora_sft.yaml
+++ b/examples/train_lora/llama3_lora_sft.yaml
@@ -10,7 +10,7 @@ lora_target: all
 ### dataset
 dataset: identity,alpaca_en_demo
 template: llama3
-cutoff_len: 1024
+cutoff_len: 2048
 max_samples: 1000
 overwrite_cache: true
 preprocessing_num_workers: 16
--- a/examples/train_lora/llama3_lora_sft_ds0.yaml
+++ b/examples/train_lora/llama3_lora_sft_ds0.yaml
@@ -11,7 +11,7 @@ deepspeed: examples/deepspeed/ds_z0_config.json
 ### dataset
 dataset: identity,alpaca_en_demo
 template: llama3
-cutoff_len: 1024
+cutoff_len: 2048
 max_samples: 1000
 overwrite_cache: true
 preprocessing_num_workers: 16
--- a/examples/train_lora/llama3_lora_sft_ds3.yaml
+++ b/examples/train_lora/llama3_lora_sft_ds3.yaml
@@ -11,7 +11,7 @@ deepspeed: examples/deepspeed/ds_z3_config.json
 ### dataset
 dataset: identity,alpaca_en_demo
 template: llama3
-cutoff_len: 1024
+cutoff_len: 2048
 max_samples: 1000
 overwrite_cache: true
 preprocessing_num_workers: 16
--- a/examples/train_lora/llama3_preprocess.yaml
+++ b/examples/train_lora/llama3_preprocess.yaml
@@ -10,7 +10,7 @@ lora_target: all
 ### dataset
 dataset: identity,alpaca_en_demo
 template: llama3
-cutoff_len: 1024
+cutoff_len: 2048
 max_samples: 1000
 overwrite_cache: true
 preprocessing_num_workers: 16
--- a/examples/train_lora/llava1_5_lora_sft.yaml
+++ b/examples/train_lora/llava1_5_lora_sft.yaml
@@ -10,7 +10,7 @@ lora_target: all
 ### dataset
 dataset: mllm_demo
 template: llava
-cutoff_len: 1024
+cutoff_len: 2048
 max_samples: 1000
 overwrite_cache: true
 preprocessing_num_workers: 16
--- a/examples/train_lora/qwen2vl_lora_dpo.yaml
+++ b/examples/train_lora/qwen2vl_lora_dpo.yaml
@@ -12,7 +12,7 @@ pref_loss: sigmoid  # choices: [sigmoid (dpo), orpo, simpo]
 ### dataset
 dataset: rlhf_v
 template: qwen2_vl
-cutoff_len: 1024
+cutoff_len: 2048
 max_samples: 1000
 overwrite_cache: true
 preprocessing_num_workers: 16
--- a/examples/train_lora/qwen2vl_lora_sft.yaml
+++ b/examples/train_lora/qwen2vl_lora_sft.yaml
@@ -10,7 +10,7 @@ lora_target: all
 ### dataset
 dataset: mllm_demo,identity  # video: mllm_video_demo
 template: qwen2_vl
-cutoff_len: 1024
+cutoff_len: 2048
 max_samples: 1000
 overwrite_cache: true
 preprocessing_num_workers: 16