added abstraction for target model backend

Author: FrankLeeeee

examples/run_llama3_eagle3_online.sh

@@ -8,15 +8,22 @@ NUM_GPUS=${1:-8}
 torchrun \
     --standalone \
     --nproc_per_node $NUM_GPUS \
-    $ROOT_DIR/scripts/train_eagle3_online.py \
+    $ROOT_DIR/scripts/train_eagle3.py \
     --target-model-path meta-llama/Meta-Llama-3.1-8B-Instruct \
     --draft-model-config $ROOT_DIR/configs/llama3-8B-eagle3.json \
-    --train-data-path $ROOT_DIR/cache/dataset/sharegpt.jsonl \
+    --train-data-path $ROOT_DIR/cache/dataset/sharegpt_train.jsonl \
     --output-dir $ROOT_DIR/outputs/llama3-8b-eagle3 \
     --num-epochs 2 \
     --batch-size 2 \
     --learning-rate 1e-4 \
     --max-length 2048 \
     --chat-template llama3 \
     --cache-dir $ROOT_DIR/cache \
-    --attention-backend flex_attention
+    --attention-backend flex_attention \
+    --target-model-backend sglang \
+    --report-to wandb \
+    --log-steps 10
+
+
+
+

requirements.txt

@@ -2,7 +2,7 @@ pre-commit
 torch==2.8.0
 torchaudio==2.8.0
 torchvision==0.23.0
-transformers==4.55.2
+transformers==4.57.1
 qwen-vl-utils==0.0.11
 datasets
 setuptools
@@ -12,5 +12,5 @@ psutil
 numpy
 accelerate
 pydantic
-sglang[all]==0.5.1
+sglang==0.5.4
 openai-harmony