423A35C7/LLaMA-Factory
1
0
Fork 0
mirror of https://github.com/hiyouga/LLaMA-Factory.git synced 2025-12-14 10:56:56 +08:00
Code Issues Packages Projects Releases Wiki Activity

add tests.cal_flops.py

Browse Source 
Former-commit-id: 469f859161
This commit is contained in:
hiyouga
2023-09-16 23:40:41 +08:00
parent 65854736c3
commit 4e6e42bb55
2 changed files with 46 additions and 4 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
src/llmtuner/tuner/core/loader.py
View File

@@ -86,10 +86,8 @@ def load_model_and_tokenizer(
# Fix config (for Qwen)
if is_trainable and hasattr(config, "fp16") and hasattr(config, "bf16"):
if model_args.compute_dtype == torch.bfloat16:
setattr(config, "bf16", True)
else:
setattr(config, "fp16", True)
setattr(config, "fp16", model_args.compute_dtype == torch.float16)
setattr(config, "bf16", model_args.compute_dtype == torch.bfloat16)
# Set RoPE scaling
if model_args.rope_scaling is not None: