Running Tasks

LlamaFactory Adapter Plugin

Configuration flow

Before running a training task, configure these environment variables in bash:

Environment variable	Description
ECO_GRPC_ADDR	Server address
ECO_CLIENT_ID	Unique user identifier [ID]
ECO_API_KEY	API authentication value [PASSWORD]
ECO_TLS_ROOT_CA	TLS root certificate path

Example:

export ECO_GRPC_ADDR="121.41.XXX.XX:80"
export ECO_CLIENT_ID="User_XX"
export ECO_API_KEY="XXXX"
export ECO_TLS_ROOT_CA="xx/rootCA.pem"

---

Usage example

Replace these values based on your environment:

• Server address
• Certificate path
• Data path
• Model path
• Configuration file path

Example command:

export ECO_GRPC_ADDR="121.41.XXX.XX:80" && \
export ECO_CLIENT_ID="User_XX" && \
export ECO_API_KEY="XXXX" && \
export ECO_TLS_ROOT_CA="/root/rootCA.pem" && \
CUDA_VISIBLE_DEVICES=0,1 accelerate launch --config_file fsdp_config.yaml \
--main_process_port 29501 src/train.py emotion_rec_sft_full_eco.yaml

---

Plugin log reference

(1) Plugin imported and initialized successfully

The log includes:

[EcoPhase] EcoMonitor initialized.

---

(2) Plugin enabled

The log includes:

[EcoPhase] API is enabled.

---

(3) Plugin inactive

The log includes:

[EcoPhase] API is disabled.

---

(4) Early stop triggered

The system automatically saves the model and prints a training summary, for example:

Task early stopped at step 200/2000. Reduction: 90.0%. Saved GPU-hours: 1.03.

This means:

Field	Meaning
200/2000	The task stopped early at step 200 out of the planned 2000 steps
Reduction: 90.0%	Training steps were reduced by about 90.0%
Saved GPU-hours: 1.03	Estimated savings of 1.03 GPU-hours

---

Notes

1. Environment variables must be configured correctly and must not contain extra spaces.
2. The root certificate must exist and be valid.
3. The plugin initialization code must be inserted at the correct position in trainer.py.
4. Confirm this before running:

enabled=True

5. Check these states first in the logs:

API is enabled
API is disabled

6. The plugin currently supports data parallelism only. Other parallel modes are not supported yet.