new evals_per_epoch and saves_per_epoch to make things cleaner (#944)

* new evals_per_epoch and saves_per_epoch to make things cleaner * update per PR feedback
axolotl-ai-cloud · Dec 12, 2023 · 5f79b82 · 5f79b82
1 parent f1de29d
commit 5f79b82
Show file tree

Hide file tree

Showing 37 changed files with 102 additions and 70 deletions.
diff --git a/README.md b/README.md
@@ -691,9 +691,11 @@ warmup_ratio: 0.05  # cannot use with warmup_steps
 learning_rate: 0.00003
 lr_quadratic_warmup:
 logging_steps:
+eval_steps: # Leave empty to eval at each epoch, integers for every N steps. decimal for fraction of total steps
+evals_per_epoch: # number of times per epoch to run evals, mutually exclusive with eval_steps
 save_strategy: # Set to `no` to skip checkpoint saves
 save_steps: # Leave empty to save at each epoch
-eval_steps: # Leave empty to eval at each epoch, integers for every N steps. decimal for fraction of total steps
+saves_per_epoch: # number of times per epoch to save a checkpoint, mutually exclusive with save_steps
 save_total_limit: # Checkpoints saved at a time
 # Maximum number of iterations to train for. It precedes num_epochs which means that
 # if both are set, num_epochs will not be guaranteed.

diff --git a/examples/cerebras/btlm-ft.yml b/examples/cerebras/btlm-ft.yml
@@ -72,8 +72,8 @@ gptq_groupsize:
 gptq_model_v1:
 
 warmup_steps: 32
-eval_steps:
-save_steps:
+evals_per_epoch: 4
+saves_per_epoch: 1
 save_total_limit:
 
 debug:

diff --git a/examples/cerebras/qlora.yml b/examples/cerebras/qlora.yml
@@ -49,8 +49,8 @@ flash_attention:
 gptq_groupsize:
 gptq_model_v1:
 warmup_steps: 10
-eval_steps: 0.05
-save_steps:
+evals_per_epoch: 4
+saves_per_epoch: 1
 debug:
 deepspeed:
 weight_decay: 0.1

diff --git a/examples/code-llama/13b/lora.yml b/examples/code-llama/13b/lora.yml
@@ -54,8 +54,8 @@ xformers_attention:
 flash_attention: true
 
 warmup_steps: 10
-eval_steps: 0.05
-save_steps:
+evals_per_epoch: 4
+saves_per_epoch: 1
 debug:
 deepspeed:
 weight_decay: 0.0

diff --git a/examples/code-llama/13b/qlora.yml b/examples/code-llama/13b/qlora.yml
@@ -56,8 +56,8 @@ xformers_attention:
 flash_attention: true
 
 warmup_steps: 10
-eval_steps: 0.05
-save_steps:
+evals_per_epoch: 4
+saves_per_epoch: 1
 debug:
 deepspeed:
 weight_decay: 0.0

diff --git a/examples/code-llama/34b/lora.yml b/examples/code-llama/34b/lora.yml
@@ -54,8 +54,8 @@ xformers_attention:
 flash_attention: true
 
 warmup_steps: 10
-eval_steps: 0.05
-save_steps:
+evals_per_epoch: 4
+saves_per_epoch: 1
 debug:
 deepspeed:
 weight_decay: 0.0

diff --git a/examples/code-llama/34b/qlora.yml b/examples/code-llama/34b/qlora.yml
@@ -56,8 +56,8 @@ xformers_attention:
 flash_attention: true
 
 warmup_steps: 10
-eval_steps: 0.05
-save_steps:
+evals_per_epoch: 4
+saves_per_epoch: 1
 debug:
 deepspeed:
 weight_decay: 0.0

diff --git a/examples/code-llama/7b/lora.yml b/examples/code-llama/7b/lora.yml
@@ -54,8 +54,8 @@ xformers_attention:
 flash_attention: true
 
 warmup_steps: 10
-eval_steps: 0.05
-save_steps:
+evals_per_epoch: 4
+saves_per_epoch: 1
 debug:
 deepspeed:
 weight_decay: 0.0

diff --git a/examples/code-llama/7b/qlora.yml b/examples/code-llama/7b/qlora.yml
@@ -56,8 +56,8 @@ xformers_attention:
 flash_attention: true
 
 warmup_steps: 10
-eval_steps: 0.05
-save_steps:
+evals_per_epoch: 4
+saves_per_epoch: 1
 debug:
 deepspeed:
 weight_decay: 0.0

diff --git a/examples/falcon/config-7b-lora.yml b/examples/falcon/config-7b-lora.yml
@@ -51,8 +51,8 @@ flash_attention:
 gptq_groupsize:
 gptq_model_v1:
 warmup_steps: 40
-eval_steps: 5
-save_steps: 43
+evals_per_epoch: 4
+saves_per_epoch: 1
 debug:
 deepspeed:
 weight_decay: 0.0

diff --git a/examples/falcon/config-7b-qlora.yml b/examples/falcon/config-7b-qlora.yml
@@ -80,8 +80,8 @@ flash_attention:
 gptq_groupsize:
 gptq_model_v1:
 warmup_steps: 10
-eval_steps: 5
-save_steps: 10
+evals_per_epoch: 4
+saves_per_epoch: 1
 debug:
 deepspeed:
 weight_decay: 0.000001

diff --git a/examples/falcon/config-7b.yml b/examples/falcon/config-7b.yml
@@ -51,8 +51,8 @@ flash_attention:
 gptq_groupsize:
 gptq_model_v1:
 warmup_steps: 40
-eval_steps: 5
-save_steps: 43
+evals_per_epoch: 4
+saves_per_epoch: 1
 debug:
 deepspeed:
 weight_decay: 0.0

diff --git a/examples/gptj/qlora.yml b/examples/gptj/qlora.yml
@@ -46,8 +46,8 @@ flash_attention:
 gptq_groupsize:
 gptq_model_v1:
 warmup_steps: 10
-eval_steps: 0.05
-save_steps:
+evals_per_epoch: 4
+saves_per_epoch: 1
 debug:
 deepspeed:
 weight_decay: 0.1

diff --git a/examples/jeopardy-bot/config.yml b/examples/jeopardy-bot/config.yml
@@ -42,8 +42,8 @@ flash_attention:
 gptq_groupsize:
 gptq_model_v1:
 warmup_steps: 20
-eval_steps: 110
-save_steps: 660
+evals_per_epoch: 4
+saves_per_epoch: 1
 debug:
 deepspeed:
 weight_decay: 0.1

diff --git a/examples/llama-2/fft_optimized.yml b/examples/llama-2/fft_optimized.yml
@@ -58,9 +58,9 @@ flash_attn_fuse_qkv: false
 flash_attn_fuse_mlp: true
 
 warmup_steps: 100
-eval_steps: 0.05
+evals_per_epoch: 4
 eval_table_size:
-save_steps:
+saves_per_epoch: 1
 debug:
 deepspeed: #deepspeed/zero2.json # multi-gpu only
 weight_decay: 0.1

diff --git a/examples/llama-2/gptq-lora.yml b/examples/llama-2/gptq-lora.yml
@@ -62,8 +62,8 @@ flash_attention:
 sdp_attention:
 flash_optimum:
 warmup_steps: 100
-eval_steps:
-save_steps:
+evals_per_epoch: 4
+saves_per_epoch: 1
 debug:
 deepspeed:
 weight_decay: 0.1

diff --git a/examples/llama-2/lora.yml b/examples/llama-2/lora.yml
@@ -54,10 +54,10 @@ xformers_attention:
 flash_attention: true
 
 warmup_steps: 10
-eval_steps: 0.05
+evals_per_epoch: 4
 eval_table_size:
 eval_table_max_new_tokens: 128
-save_steps:
+saves_per_epoch: 1
 debug:
 deepspeed:
 weight_decay: 0.0

diff --git a/examples/llama-2/qlora.yml b/examples/llama-2/qlora.yml
@@ -56,9 +56,9 @@ xformers_attention:
 flash_attention: true
 
 warmup_steps: 10
-eval_steps: 0.05
+evals_per_epoch: 4
 eval_table_size:
-save_steps:
+saves_per_epoch: 1
 debug:
 deepspeed:
 weight_decay: 0.0

diff --git a/examples/llama-2/relora.yml b/examples/llama-2/relora.yml
@@ -60,8 +60,8 @@ xformers_attention:
 flash_attention: true
 
 warmup_steps: 10
-eval_steps: 0.05
-save_steps: 50
+evals_per_epoch: 4
+saves_per_epoch: 1
 debug:
 deepspeed:
 weight_decay: 0.0

diff --git a/examples/llama-2/tiny-llama.yml b/examples/llama-2/tiny-llama.yml
@@ -54,9 +54,9 @@ xformers_attention:
 flash_attention: true
 
 warmup_steps: 10
-eval_steps: 0.05
+evals_per_epoch: 4
 eval_table_size:
-save_steps:
+saves_per_epoch: 1
 debug:
 deepspeed:
 weight_decay: 0.0

diff --git a/examples/mamba/config.yml b/examples/mamba/config.yml
@@ -47,10 +47,10 @@ xformers_attention:
 flash_attention:
 
 warmup_steps: 10
-eval_steps:
+evals_per_epoch: 4
 eval_table_size:
 eval_table_max_new_tokens: 128
-save_steps:  0.25
+saves_per_epoch: 1
 debug:
 deepspeed:
 weight_decay: 0.0

diff --git a/examples/mistral/config.yml b/examples/mistral/config.yml
@@ -46,10 +46,10 @@ xformers_attention:
 flash_attention: true
 
 warmup_steps: 10
-eval_steps: 0.05
+evals_per_epoch: 4
 eval_table_size:
 eval_table_max_new_tokens: 128
-save_steps:
+saves_per_epoch: 1
 debug:
 deepspeed:
 weight_decay: 0.0

diff --git a/examples/mistral/mixtral.yml b/examples/mistral/mixtral.yml
@@ -67,10 +67,10 @@ loss_watchdog_threshold: 5.0
 loss_watchdog_patience: 3
 
 warmup_steps: 10
-eval_steps:
+evals_per_epoch: 4
 eval_table_size:
 eval_table_max_new_tokens: 128
-save_steps:
+saves_per_epoch: 1
 debug:
 deepspeed: deepspeed/zero2.json
 weight_decay: 0.0

diff --git a/examples/mistral/qlora.yml b/examples/mistral/qlora.yml
@@ -66,10 +66,10 @@ loss_watchdog_threshold: 5.0
 loss_watchdog_patience: 3
 
 warmup_steps: 10
-eval_steps: 0.05
+evals_per_epoch: 4
 eval_table_size:
 eval_table_max_new_tokens: 128
-save_steps:
+saves_per_epoch: 1
 debug:
 deepspeed:
 weight_decay: 0.0

diff --git a/examples/mpt-7b/config.yml b/examples/mpt-7b/config.yml
@@ -44,8 +44,8 @@ flash_attention:
 gptq_groupsize:
 gptq_model_v1:
 warmup_steps: 20
-eval_steps: 110
-save_steps: 660
+evals_per_epoch: 4
+saves_per_epoch: 1
 debug:
 deepspeed:
 weight_decay: 0.0001

diff --git a/examples/openllama-3b/config.yml b/examples/openllama-3b/config.yml
@@ -49,8 +49,8 @@ flash_attention: true
 gptq_groupsize:
 gptq_model_v1:
 warmup_steps: 20
-eval_steps: 0.05
-save_steps:
+evals_per_epoch: 4
+saves_per_epoch: 1
 debug:
 deepspeed:
 weight_decay: 0.1

diff --git a/examples/openllama-3b/lora.yml b/examples/openllama-3b/lora.yml
@@ -54,8 +54,8 @@ flash_attention: true
 gptq_groupsize:
 gptq_model_v1:
 warmup_steps: 20
-eval_steps: 0.05
-save_steps:
+evals_per_epoch: 4
+saves_per_epoch: 1
 debug:
 deepspeed:
 weight_decay: 0.1

diff --git a/examples/openllama-3b/qlora.yml b/examples/openllama-3b/qlora.yml
@@ -48,8 +48,8 @@ flash_attention: true
 gptq_groupsize:
 gptq_model_v1:
 warmup_steps: 20
-eval_steps: 0.05
-save_steps:
+evals_per_epoch: 4
+saves_per_epoch: 1
 debug:
 deepspeed:
 weight_decay: 0.1

diff --git a/examples/phi/phi-ft.yml b/examples/phi/phi-ft.yml
@@ -59,8 +59,8 @@ xformers_attention:
 flash_attention:
 
 warmup_steps: 100
-eval_steps: 0.05
-save_steps:
+evals_per_epoch: 4
+saves_per_epoch: 1
 debug:
 deepspeed:
 weight_decay: 0.1

diff --git a/examples/phi/phi-qlora.yml b/examples/phi/phi-qlora.yml
@@ -59,8 +59,8 @@ xformers_attention:
 flash_attention:
 
 warmup_steps: 100
-eval_steps: 0.05
-save_steps:
+evals_per_epoch: 4
+saves_per_epoch: 1
 debug:
 deepspeed:
 weight_decay: 0.1

diff --git a/examples/pythia/lora.yml b/examples/pythia/lora.yml
@@ -33,5 +33,5 @@ early_stopping_patience:
 resume_from_checkpoint:
 local_rank:
 weight_decay: 0.1
-eval_steps: 0.05
+evals_per_epoch: 4
 logging_steps: 1
diff --git a/examples/qwen/lora.yml b/examples/qwen/lora.yml
@@ -56,10 +56,10 @@ xformers_attention:
 flash_attention:
 
 warmup_steps: 10
-eval_steps: 0.05
+evals_per_epoch: 4
 eval_table_size:
 eval_table_max_new_tokens: 128
-save_steps:
+saves_per_epoch: 1
 debug:
 deepspeed:
 weight_decay: 0.0

diff --git a/examples/qwen/qlora.yml b/examples/qwen/qlora.yml
@@ -56,10 +56,10 @@ xformers_attention:
 flash_attention:
 
 warmup_steps: 10
-eval_steps: 0.05
+evals_per_epoch: 4
 eval_table_size:
 eval_table_max_new_tokens: 128
-save_steps:
+saves_per_epoch: 1
 debug:
 deepspeed:
 weight_decay: 0.0

diff --git a/examples/redpajama/config-3b.yml b/examples/redpajama/config-3b.yml
@@ -45,8 +45,8 @@ flash_attention:
 gptq_groupsize:
 gptq_model_v1:
 warmup_steps: 20
-eval_steps: 110
-save_steps: 660
+evals_per_epoch: 4
+saves_per_epoch: 1
 debug:
 deepspeed:
 weight_decay: 0.0001

diff --git a/examples/replit-3b/config-lora.yml b/examples/replit-3b/config-lora.yml
@@ -45,8 +45,8 @@ flash_attention:
 gptq_groupsize:
 gptq_model_v1:
 warmup_steps: 20
-eval_steps: 50
-save_steps:
+evals_per_epoch: 4
+saves_per_epoch: 1
 debug:
 deepspeed:
 weight_decay: 0