cmedina/agenx-lora-training
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

fix: use FP8 model with 32GB CPU offload

Browse Source
This commit is contained in:
Christian Medina
2026-07-01 07:29:32 -04:00
parent 9d92531c55
commit 6f631f00ae
1 changed files with 4 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files

5
training/configs/ornith-35b-lora.yaml
View File

@@ -1,7 +1,7 @@
# LoRA Training Configuration for Llama-2-7b
# Dataset: cyron_summary_lora_dataset (20k examples)
base_model: /data/models/Ornith-1.0-35B
base_model: /data/models/Ornith-1.0-35B-FP8
model_type: LlamaForCausalLM
tokenizer_type: LlamaTokenizer
@@ -59,9 +59,12 @@ deepspeed_config:
offload_param:
device: cpu
pin_memory: true
offload_params_device: cpu
gradient_clipping: 1.0
train_batch_size: auto
train_micro_batch_size_per_gpu: auto
# Offload 32GB to RAM
zero_hierarchical_offload: true
# Evaluation
eval_strategy: steps