tocmo0nlord/axolotl
1
0
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases Wiki Activity

fix deprecated type:sharegpt and flash_attention config keys

Browse Source
This commit is contained in:
tocmo0nlord
2026-05-13 12:52:25 +00:00
parent e9a3fd483f
commit b9ceebfe7e
1 changed files with 3 additions and 5 deletions
Download Patch File Download Diff File Expand all files Collapse all files

8
human_chat_qlora.yml
View File

@@ -33,13 +33,11 @@ default_system_message: >-
# OpenHermes-2.5: broad instruction coverage — sampled to 5% to keep balance # OpenHermes-2.5: broad instruction coverage — sampled to 5% to keep balance
datasets: datasets:
- path: LDJnr/Capybara - path: LDJnr/Capybara
type: sharegpt type: chat_template
conversation: llama3
split: train split: train
- path: teknium/OpenHermes-2.5 - path: teknium/OpenHermes-2.5
type: sharegpt type: chat_template
conversation: llama3
split: "train[:5%]" split: "train[:5%]"
dataset_prepared_path: last_run_prepared dataset_prepared_path: last_run_prepared
@@ -75,7 +73,7 @@ tf32: false
# --- Memory & speed --- # --- Memory & speed ---
gradient_checkpointing: true gradient_checkpointing: true
flash_attention: true attn_implementation: flash_attention_2
# --- Logging & checkpointing --- # --- Logging & checkpointing ---
logging_steps: 10 logging_steps: 10