simplify

Dockerfile

@@ -18,12 +18,6 @@ COPY . .
 RUN pip install --upgrade pip && \
     pip install --no-cache-dir -r requirements.txt
 
-# Make script executable and move it to PATH
-RUN install -m 755 sft.py /usr/local/bin/sft
-
-# Set environment variable for Hugging Face cache
-# ENV HF_HOME=/usr/src/app/.cache/huggingface
-
 # Make /usr/src/app/ a writable directory
 RUN chmod -R 777 $HOME
 

configs/Qwen3-4B-Base_a10g-small.yaml

@@ -1,43 +0,0 @@
-# Model arguments
-model_name_or_path: Qwen/Qwen2.5-1.5B
-model_revision: main
-torch_dtype: bfloat16
-
-# Data training arguments
-dataset_name: trl-lib/tldr
-dataset_num_proc: 4
-
-# SFT trainer config
-bf16: true
-do_eval: false
-eval_strategy: 'no'
-gradient_accumulation_steps: 8
-gradient_checkpointing: true
-gradient_checkpointing_kwargs:
-  use_reentrant: false
-# hub_model_id: open-r1/OlympicCoder-7B
-hub_strategy: every_save
-learning_rate: 1.0e-05
-log_level: info
-logging_steps: 1
-logging_strategy: steps
-lr_scheduler_type: cosine_with_min_lr
-lr_scheduler_kwargs:
-  min_lr_rate: 0.1
-packing: false
-max_grad_norm: 0.2
-max_length: 512
-max_steps: -1
-num_train_epochs: 10
-output_dir: data/Qwen2.5-0.5B-SFT
-overwrite_output_dir: true
-per_device_eval_batch_size: 1
-per_device_train_batch_size: 2
-push_to_hub: true
-report_to:
-- wandb
-save_strategy: epoch
-save_total_limit: 1
-seed: 42
-use_liger_kernel: true
-warmup_ratio: 0.03

requirements.txt

@@ -1,3 +1,3 @@
 git+https://github.com/huggingface/trl.git@main#egg=trl
-gradio
-liger-kernel
+liger-kernel
+gradio

sft.py

@@ -1,24 +0,0 @@
-#!/usr/bin/env python3
-import argparse
-import subprocess
-import os
-
-parser = argparse.ArgumentParser(description="Demo script for the model.")
-
-parser.add_argument("--model", type=str)
-parser.add_argument("--dataset", type=str)
-parser.add_argument("--flavor", type=str)
-parser.add_argument("--token", type=str)
-
-# Parse the command line arguments
-args = parser.parse_args()
-
-# Get the config file based on the model and flavor
-config_file = f"configs/{args.model}_{args.flavor}.yaml"
-
-# Check if the config file exists
-if not os.path.exists(config_file):
-    raise RuntimeError(f"Training model {args.model} with flavor {args.flavor} is not supported.")
-
-# Run another Python script
-subprocess.run(["trl", "sft", "--config", config_file, "--dataset_name", args.dataset])